diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 47c6c0f754f..988000c6401 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -1,17 +1,17 @@ { "sample_spec": { - "max_tokens": "LEGACY parameter. set to max_output_tokens if provider specifies it. IF not set to max_input_tokens, if provider specifies it.", + "max_tokens": "LEGACY parameter. set to max_output_tokens if provider specifies it. IF not set to max_input_tokens, if provider specifies it.", "max_input_tokens": "max input tokens, if the provider specifies it. if not default to max_tokens", - "max_output_tokens": "max output tokens, if the provider specifies it. if not default to max_tokens", - "input_cost_per_token": 0.0000, - "output_cost_per_token": 0.000, - "output_cost_per_reasoning_token": 0.000, + "max_output_tokens": "max output tokens, if the provider specifies it. if not default to max_tokens", + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, + "output_cost_per_reasoning_token": 0.0, "litellm_provider": "one of https://docs.litellm.ai/docs/providers", "mode": "one of: chat, embedding, completion, image_generation, audio_transcription, audio_speech, image_generation, moderation, rerank", "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_vision": true, - "supports_audio_input": true, + "supports_audio_input": true, "supports_audio_output": true, "supports_prompt_caching": true, "supports_response_schema": true, @@ -19,16 +19,16 @@ "supports_reasoning": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 0.0000, - "search_context_size_medium": 0.0000, - "search_context_size_high": 0.0000 + "search_context_size_low": 0.0, + "search_context_size_medium": 0.0, + "search_context_size_high": 0.0 }, "deprecation_date": "date when the model becomes deprecated in the format YYYY-MM-DD" }, "omni-moderation-latest": { "max_tokens": 32768, "max_input_tokens": 32768, - "max_output_tokens": 0, + "max_output_tokens": 0, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "openai", @@ -37,7 +37,7 @@ "omni-moderation-latest-intents": { "max_tokens": 32768, "max_input_tokens": 32768, - "max_output_tokens": 0, + "max_output_tokens": 0, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "openai", @@ -46,18 +46,18 @@ "omni-moderation-2024-09-26": { "max_tokens": 32768, "max_input_tokens": 32768, - "max_output_tokens": 0, + "max_output_tokens": 0, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "openai", "mode": "moderation" }, "gpt-4": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 8192, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.00006, + "max_output_tokens": 4096, + "input_cost_per_token": 3e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -69,16 +69,25 @@ "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 2e-6, - "output_cost_per_token": 8e-6, - "input_cost_per_token_batches": 1e-6, - "output_cost_per_token_batches": 4e-6, - "cache_read_input_token_cost": 0.5e-6, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, + "input_cost_per_token_batches": 1e-06, + "output_cost_per_token_batches": 4e-06, + "cache_read_input_token_cost": 5e-07, "litellm_provider": "openai", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_pdf_input": true, "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -90,25 +99,34 @@ "supports_native_streaming": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 30e-3, - "search_context_size_medium": 35e-3, - "search_context_size_high": 50e-3 + "search_context_size_low": 0.03, + "search_context_size_medium": 0.035, + "search_context_size_high": 0.05 } }, "gpt-4.1-2025-04-14": { "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 2e-6, - "output_cost_per_token": 8e-6, - "input_cost_per_token_batches": 1e-6, - "output_cost_per_token_batches": 4e-6, - "cache_read_input_token_cost": 0.5e-6, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, + "input_cost_per_token_batches": 1e-06, + "output_cost_per_token_batches": 4e-06, + "cache_read_input_token_cost": 5e-07, "litellm_provider": "openai", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_pdf_input": true, "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -120,25 +138,34 @@ "supports_native_streaming": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 30e-3, - "search_context_size_medium": 35e-3, - "search_context_size_high": 50e-3 + "search_context_size_low": 0.03, + "search_context_size_medium": 0.035, + "search_context_size_high": 0.05 } }, "gpt-4.1-mini": { "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 0.4e-6, - "output_cost_per_token": 1.6e-6, - "input_cost_per_token_batches": 0.2e-6, - "output_cost_per_token_batches": 0.8e-6, - "cache_read_input_token_cost": 0.1e-6, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 1.6e-06, + "input_cost_per_token_batches": 2e-07, + "output_cost_per_token_batches": 8e-07, + "cache_read_input_token_cost": 1e-07, "litellm_provider": "openai", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_pdf_input": true, "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -150,25 +177,34 @@ "supports_native_streaming": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 25e-3, - "search_context_size_medium": 27.5e-3, - "search_context_size_high": 30e-3 + "search_context_size_low": 0.025, + "search_context_size_medium": 0.0275, + "search_context_size_high": 0.03 } }, "gpt-4.1-mini-2025-04-14": { "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 0.4e-6, - "output_cost_per_token": 1.6e-6, - "input_cost_per_token_batches": 0.2e-6, - "output_cost_per_token_batches": 0.8e-6, - "cache_read_input_token_cost": 0.1e-6, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 1.6e-06, + "input_cost_per_token_batches": 2e-07, + "output_cost_per_token_batches": 8e-07, + "cache_read_input_token_cost": 1e-07, "litellm_provider": "openai", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_pdf_input": true, "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -180,25 +216,34 @@ "supports_native_streaming": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 25e-3, - "search_context_size_medium": 27.5e-3, - "search_context_size_high": 30e-3 + "search_context_size_low": 0.025, + "search_context_size_medium": 0.0275, + "search_context_size_high": 0.03 } }, "gpt-4.1-nano": { "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 0.1e-6, - "output_cost_per_token": 0.4e-6, - "input_cost_per_token_batches": 0.05e-6, - "output_cost_per_token_batches": 0.2e-6, - "cache_read_input_token_cost": 0.025e-6, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, + "input_cost_per_token_batches": 5e-08, + "output_cost_per_token_batches": 2e-07, + "cache_read_input_token_cost": 2.5e-08, "litellm_provider": "openai", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_pdf_input": true, "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -213,16 +258,25 @@ "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 0.1e-6, - "output_cost_per_token": 0.4e-6, - "input_cost_per_token_batches": 0.05e-6, - "output_cost_per_token_batches": 0.2e-6, - "cache_read_input_token_cost": 0.025e-6, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, + "input_cost_per_token_batches": 5e-08, + "output_cost_per_token_batches": 2e-07, + "cache_read_input_token_cost": 2.5e-08, "litellm_provider": "openai", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_pdf_input": true, "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -237,11 +291,11 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000010, - "input_cost_per_token_batches": 0.00000125, - "output_cost_per_token_batches": 0.00000500, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "input_cost_per_token_batches": 1.25e-06, + "output_cost_per_token_batches": 5e-06, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -254,38 +308,38 @@ "supports_tool_choice": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 0.030, + "search_context_size_low": 0.03, "search_context_size_medium": 0.035, - "search_context_size_high": 0.050 + "search_context_size_high": 0.05 } }, "watsonx/ibm/granite-3-8b-instruct": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 1024, - "input_cost_per_token": 0.0002, - "output_cost_per_token": 0.0002, - "litellm_provider": "watsonx", - "mode": "chat", - "supports_function_calling": true, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 1024, + "input_cost_per_token": 0.0002, + "output_cost_per_token": 0.0002, + "litellm_provider": "watsonx", + "mode": "chat", + "supports_function_calling": true, "supports_tool_choice": true, - "supports_parallel_function_calling": false, - "supports_vision": false, - "supports_audio_input": false, - "supports_audio_output": false, - "supports_prompt_caching": true, - "supports_response_schema": true, + "supports_parallel_function_calling": false, + "supports_vision": false, + "supports_audio_input": false, + "supports_audio_output": false, + "supports_prompt_caching": true, + "supports_response_schema": true, "supports_system_messages": true }, "gpt-4o-search-preview-2025-03-11": { "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000010, - "input_cost_per_token_batches": 0.00000125, - "output_cost_per_token_batches": 0.00000500, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "input_cost_per_token_batches": 1.25e-06, + "output_cost_per_token_batches": 5e-06, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -298,20 +352,20 @@ "supports_tool_choice": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 0.030, + "search_context_size_low": 0.03, "search_context_size_medium": 0.035, - "search_context_size_high": 0.050 + "search_context_size_high": 0.05 } - }, + }, "gpt-4o-search-preview": { - "max_tokens": 16384, + "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000010, - "input_cost_per_token_batches": 0.00000125, - "output_cost_per_token_batches": 0.00000500, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "input_cost_per_token_batches": 1.25e-06, + "output_cost_per_token_batches": 5e-06, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -324,20 +378,20 @@ "supports_tool_choice": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 0.030, + "search_context_size_low": 0.03, "search_context_size_medium": 0.035, - "search_context_size_high": 0.050 + "search_context_size_high": 0.05 } }, "gpt-4.5-preview": { "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.000075, + "input_cost_per_token": 7.5e-05, "output_cost_per_token": 0.00015, - "input_cost_per_token_batches": 0.0000375, - "output_cost_per_token_batches": 0.000075, - "cache_read_input_token_cost": 0.0000375, + "input_cost_per_token_batches": 3.75e-05, + "output_cost_per_token_batches": 7.5e-05, + "cache_read_input_token_cost": 3.75e-05, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -353,11 +407,11 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.000075, + "input_cost_per_token": 7.5e-05, "output_cost_per_token": 0.00015, - "input_cost_per_token_batches": 0.0000375, - "output_cost_per_token_batches": 0.000075, - "cache_read_input_token_cost": 0.0000375, + "input_cost_per_token_batches": 3.75e-05, + "output_cost_per_token_batches": 7.5e-05, + "cache_read_input_token_cost": 3.75e-05, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -374,9 +428,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, + "input_cost_per_token": 2.5e-06, "input_cost_per_audio_token": 0.0001, - "output_cost_per_token": 0.000010, + "output_cost_per_token": 1e-05, "output_cost_per_audio_token": 0.0002, "litellm_provider": "openai", "mode": "chat", @@ -391,10 +445,10 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "input_cost_per_audio_token": 0.00004, - "output_cost_per_token": 0.000010, - "output_cost_per_audio_token": 0.00008, + "input_cost_per_token": 2.5e-06, + "input_cost_per_audio_token": 4e-05, + "output_cost_per_token": 1e-05, + "output_cost_per_audio_token": 8e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -408,9 +462,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, + "input_cost_per_token": 2.5e-06, "input_cost_per_audio_token": 0.0001, - "output_cost_per_token": 0.000010, + "output_cost_per_token": 1e-05, "output_cost_per_audio_token": 0.0002, "litellm_provider": "openai", "mode": "chat", @@ -425,10 +479,10 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "input_cost_per_audio_token": 0.00001, - "output_cost_per_token": 0.0000006, - "output_cost_per_audio_token": 0.00002, + "input_cost_per_token": 1.5e-07, + "input_cost_per_audio_token": 1e-05, + "output_cost_per_token": 6e-07, + "output_cost_per_audio_token": 2e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -442,10 +496,10 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "input_cost_per_audio_token": 0.00001, - "output_cost_per_token": 0.0000006, - "output_cost_per_audio_token": 0.00002, + "input_cost_per_token": 1.5e-07, + "input_cost_per_audio_token": 1e-05, + "output_cost_per_token": 6e-07, + "output_cost_per_audio_token": 2e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -459,37 +513,11 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000060, - "input_cost_per_token_batches": 0.000000075, - "output_cost_per_token_batches": 0.00000030, - "cache_read_input_token_cost": 0.000000075, - "litellm_provider": "openai", - "mode": "chat", - "supports_pdf_input": true, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_prompt_caching": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_web_search": true, - "search_context_cost_per_query": { - "search_context_size_low": 0.025, - "search_context_size_medium": 0.0275, - "search_context_size_high": 0.030 - } - }, - "gpt-4o-mini-search-preview-2025-03-11":{ - "max_tokens": 16384, - "max_input_tokens": 128000, - "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000060, - "input_cost_per_token_batches": 0.000000075, - "output_cost_per_token_batches": 0.00000030, - "cache_read_input_token_cost": 0.000000075, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "input_cost_per_token_batches": 7.5e-08, + "output_cost_per_token_batches": 3e-07, + "cache_read_input_token_cost": 7.5e-08, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -504,18 +532,44 @@ "search_context_cost_per_query": { "search_context_size_low": 0.025, "search_context_size_medium": 0.0275, - "search_context_size_high": 0.030 + "search_context_size_high": 0.03 + } + }, + "gpt-4o-mini-search-preview-2025-03-11": { + "max_tokens": 16384, + "max_input_tokens": 128000, + "max_output_tokens": 16384, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "input_cost_per_token_batches": 7.5e-08, + "output_cost_per_token_batches": 3e-07, + "cache_read_input_token_cost": 7.5e-08, + "litellm_provider": "openai", + "mode": "chat", + "supports_pdf_input": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_response_schema": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_web_search": true, + "search_context_cost_per_query": { + "search_context_size_low": 0.025, + "search_context_size_medium": 0.0275, + "search_context_size_high": 0.03 } }, "gpt-4o-mini-search-preview": { "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000060, - "input_cost_per_token_batches": 0.000000075, - "output_cost_per_token_batches": 0.00000030, - "cache_read_input_token_cost": 0.000000075, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "input_cost_per_token_batches": 7.5e-08, + "output_cost_per_token_batches": 3e-07, + "cache_read_input_token_cost": 7.5e-08, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -530,18 +584,18 @@ "search_context_cost_per_query": { "search_context_size_low": 0.025, "search_context_size_medium": 0.0275, - "search_context_size_high": 0.030 + "search_context_size_high": 0.03 } }, "gpt-4o-mini-2024-07-18": { "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000060, - "input_cost_per_token_batches": 0.000000075, - "output_cost_per_token_batches": 0.00000030, - "cache_read_input_token_cost": 0.000000075, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "input_cost_per_token_batches": 7.5e-08, + "output_cost_per_token_batches": 3e-07, + "cache_read_input_token_cost": 7.5e-08, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -553,9 +607,9 @@ "supports_system_messages": true, "supports_tool_choice": true, "search_context_cost_per_query": { - "search_context_size_low": 30.00, - "search_context_size_medium": 35.00, - "search_context_size_high": 50.00 + "search_context_size_low": 30.0, + "search_context_size_medium": 35.0, + "search_context_size_high": 50.0 } }, "o1-pro": { @@ -564,7 +618,7 @@ "max_output_tokens": 100000, "input_cost_per_token": 0.00015, "output_cost_per_token": 0.0006, - "input_cost_per_token_batches": 0.000075, + "input_cost_per_token_batches": 7.5e-05, "output_cost_per_token_batches": 0.0003, "litellm_provider": "openai", "mode": "responses", @@ -578,9 +632,17 @@ "supports_tool_choice": true, "supports_native_streaming": false, "supports_reasoning": true, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], - "supported_endpoints": ["/v1/responses", "/v1/batch"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supported_endpoints": [ + "/v1/responses", + "/v1/batch" + ] }, "o1-pro-2025-03-19": { "max_tokens": 100000, @@ -588,7 +650,7 @@ "max_output_tokens": 100000, "input_cost_per_token": 0.00015, "output_cost_per_token": 0.0006, - "input_cost_per_token_batches": 0.000075, + "input_cost_per_token_batches": 7.5e-05, "output_cost_per_token_batches": 0.0003, "litellm_provider": "openai", "mode": "responses", @@ -602,17 +664,25 @@ "supports_tool_choice": true, "supports_native_streaming": false, "supports_reasoning": true, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], - "supported_endpoints": ["/v1/responses", "/v1/batch"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supported_endpoints": [ + "/v1/responses", + "/v1/batch" + ] }, "o1": { "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.00006, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -629,9 +699,9 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.0000044, - "cache_read_input_token_cost": 0.00000055, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 5.5e-07, "litellm_provider": "openai", "mode": "chat", "supports_vision": true, @@ -642,13 +712,20 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 12e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.2e-05, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -662,9 +739,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 1e-5, - "output_cost_per_token": 4e-5, - "cache_read_input_token_cost": 2.5e-6, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 4e-05, + "cache_read_input_token_cost": 2.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -680,9 +757,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 1e-5, - "output_cost_per_token": 4e-5, - "cache_read_input_token_cost": 2.5e-6, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 4e-05, + "cache_read_input_token_cost": 2.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -698,9 +775,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.0000044, - "cache_read_input_token_cost": 0.00000055, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 5.5e-07, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -715,9 +792,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.0000044, - "cache_read_input_token_cost": 0.00000055, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 5.5e-07, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -732,9 +809,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 1.1e-6, - "output_cost_per_token": 4.4e-6, - "cache_read_input_token_cost": 2.75e-7, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 2.75e-07, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -750,9 +827,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 1.1e-6, - "output_cost_per_token": 4.4e-6, - "cache_read_input_token_cost": 2.75e-7, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 2.75e-07, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -768,9 +845,9 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000012, - "cache_read_input_token_cost": 0.0000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.2e-05, + "cache_read_input_token_cost": 1.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -782,9 +859,9 @@ "max_tokens": 32768, "max_input_tokens": 128000, "max_output_tokens": 32768, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -796,9 +873,9 @@ "max_tokens": 32768, "max_input_tokens": 128000, "max_output_tokens": 32768, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -810,9 +887,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -829,8 +906,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -845,10 +922,10 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000015, - "input_cost_per_token_batches": 0.0000025, - "output_cost_per_token_batches": 0.0000075, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 1.5e-05, + "input_cost_per_token_batches": 2.5e-06, + "output_cost_per_token_batches": 7.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -863,11 +940,11 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000010, - "input_cost_per_token_batches": 0.00000125, - "output_cost_per_token_batches": 0.0000050, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "input_cost_per_token_batches": 1.25e-06, + "output_cost_per_token_batches": 5e-06, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -880,20 +957,20 @@ "supports_tool_choice": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 0.030, + "search_context_size_low": 0.03, "search_context_size_medium": 0.035, - "search_context_size_high": 0.050 + "search_context_size_high": 0.05 } }, "gpt-4o-2024-11-20": { "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000010, - "input_cost_per_token_batches": 0.00000125, - "output_cost_per_token_batches": 0.0000050, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "input_cost_per_token_batches": 1.25e-06, + "output_cost_per_token_batches": 5e-06, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -909,11 +986,11 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, + "input_cost_per_token": 5e-06, "input_cost_per_audio_token": 0.0001, - "cache_read_input_token_cost": 0.0000025, - "cache_creation_input_audio_token_cost": 0.00002, - "output_cost_per_token": 0.00002, + "cache_read_input_token_cost": 2.5e-06, + "cache_creation_input_audio_token_cost": 2e-05, + "output_cost_per_token": 2e-05, "output_cost_per_audio_token": 0.0002, "litellm_provider": "openai", "mode": "chat", @@ -928,11 +1005,11 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, - "input_cost_per_audio_token": 0.00004, - "cache_read_input_token_cost": 0.0000025, - "output_cost_per_token": 0.00002, - "output_cost_per_audio_token": 0.00008, + "input_cost_per_token": 5e-06, + "input_cost_per_audio_token": 4e-05, + "cache_read_input_token_cost": 2.5e-06, + "output_cost_per_token": 2e-05, + "output_cost_per_audio_token": 8e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -946,11 +1023,11 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, - "input_cost_per_audio_token": 0.00004, - "cache_read_input_token_cost": 0.0000025, - "output_cost_per_token": 0.00002, - "output_cost_per_audio_token": 0.00008, + "input_cost_per_token": 5e-06, + "input_cost_per_audio_token": 4e-05, + "cache_read_input_token_cost": 2.5e-06, + "output_cost_per_token": 2e-05, + "output_cost_per_audio_token": 8e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -964,12 +1041,12 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000006, - "input_cost_per_audio_token": 0.00001, - "cache_read_input_token_cost": 0.0000003, - "cache_creation_input_audio_token_cost": 0.0000003, - "output_cost_per_token": 0.0000024, - "output_cost_per_audio_token": 0.00002, + "input_cost_per_token": 6e-07, + "input_cost_per_audio_token": 1e-05, + "cache_read_input_token_cost": 3e-07, + "cache_creation_input_audio_token_cost": 3e-07, + "output_cost_per_token": 2.4e-06, + "output_cost_per_audio_token": 2e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -983,12 +1060,12 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000006, - "input_cost_per_audio_token": 0.00001, - "cache_read_input_token_cost": 0.0000003, - "cache_creation_input_audio_token_cost": 0.0000003, - "output_cost_per_token": 0.0000024, - "output_cost_per_audio_token": 0.00002, + "input_cost_per_token": 6e-07, + "input_cost_per_audio_token": 1e-05, + "cache_read_input_token_cost": 3e-07, + "cache_creation_input_audio_token_cost": 3e-07, + "output_cost_per_token": 2.4e-06, + "output_cost_per_audio_token": 2e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1002,8 +1079,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -1017,8 +1094,8 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.00006, + "input_cost_per_token": 3e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "openai", "mode": "chat", "supports_prompt_caching": true, @@ -1029,8 +1106,8 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.00006, + "input_cost_per_token": 3e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1043,7 +1120,7 @@ "max_tokens": 4096, "max_input_tokens": 32768, "max_output_tokens": 4096, - "input_cost_per_token": 0.00006, + "input_cost_per_token": 6e-05, "output_cost_per_token": 0.00012, "litellm_provider": "openai", "mode": "chat", @@ -1055,7 +1132,7 @@ "max_tokens": 4096, "max_input_tokens": 32768, "max_output_tokens": 4096, - "input_cost_per_token": 0.00006, + "input_cost_per_token": 6e-05, "output_cost_per_token": 0.00012, "litellm_provider": "openai", "mode": "chat", @@ -1067,7 +1144,7 @@ "max_tokens": 4096, "max_input_tokens": 32768, "max_output_tokens": 4096, - "input_cost_per_token": 0.00006, + "input_cost_per_token": 6e-05, "output_cost_per_token": 0.00012, "litellm_provider": "openai", "mode": "chat", @@ -1079,8 +1156,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -1095,8 +1172,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -1111,8 +1188,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1125,8 +1202,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1139,8 +1216,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "openai", "mode": "chat", "supports_vision": true, @@ -1154,8 +1231,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "openai", "mode": "chat", "supports_vision": true, @@ -1169,8 +1246,8 @@ "max_tokens": 4097, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1182,8 +1259,8 @@ "max_tokens": 4097, "max_input_tokens": 4097, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "openai", "mode": "chat", "supports_prompt_caching": true, @@ -1194,8 +1271,8 @@ "max_tokens": 4097, "max_input_tokens": 4097, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1207,8 +1284,8 @@ "max_tokens": 16385, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000010, - "output_cost_per_token": 0.0000020, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1221,8 +1298,8 @@ "max_tokens": 16385, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1235,8 +1312,8 @@ "max_tokens": 16385, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000004, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 4e-06, "litellm_provider": "openai", "mode": "chat", "supports_prompt_caching": true, @@ -1247,8 +1324,8 @@ "max_tokens": 16385, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000004, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 4e-06, "litellm_provider": "openai", "mode": "chat", "supports_prompt_caching": true, @@ -1259,10 +1336,10 @@ "max_tokens": 4096, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000006, - "input_cost_per_token_batches": 0.0000015, - "output_cost_per_token_batches": 0.000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 6e-06, + "input_cost_per_token_batches": 1.5e-06, + "output_cost_per_token_batches": 3e-06, "litellm_provider": "openai", "mode": "chat", "supports_system_messages": true, @@ -1272,8 +1349,8 @@ "max_tokens": 4096, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "openai", "mode": "chat", "supports_system_messages": true, @@ -1283,8 +1360,8 @@ "max_tokens": 4096, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "openai", "mode": "chat", "supports_system_messages": true, @@ -1294,8 +1371,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "openai", "mode": "chat", "supports_system_messages": true, @@ -1305,8 +1382,8 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.00006, + "input_cost_per_token": 3e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1318,10 +1395,10 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000375, - "output_cost_per_token": 0.000015, - "input_cost_per_token_batches": 0.000001875, - "output_cost_per_token_batches": 0.000007500, + "input_cost_per_token": 3.75e-06, + "output_cost_per_token": 1.5e-05, + "input_cost_per_token_batches": 1.875e-06, + "output_cost_per_token_batches": 7.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -1336,9 +1413,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000375, - "cache_creation_input_token_cost": 0.000001875, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3.75e-06, + "cache_creation_input_token_cost": 1.875e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -1354,11 +1431,11 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000012, - "input_cost_per_token_batches": 0.000000150, - "output_cost_per_token_batches": 0.000000600, - "cache_read_input_token_cost": 0.00000015, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 1.2e-06, + "input_cost_per_token_batches": 1.5e-07, + "output_cost_per_token_batches": 6e-07, + "cache_read_input_token_cost": 1.5e-07, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1374,10 +1451,10 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 4096, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000002, - "input_cost_per_token_batches": 0.000001, - "output_cost_per_token_batches": 0.000001, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 2e-06, + "input_cost_per_token_batches": 1e-06, + "output_cost_per_token_batches": 1e-06, "litellm_provider": "text-completion-openai", "mode": "completion" }, @@ -1385,10 +1462,10 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000004, - "output_cost_per_token": 0.0000004, - "input_cost_per_token_batches": 0.0000002, - "output_cost_per_token_batches": 0.0000002, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 4e-07, + "input_cost_per_token_batches": 2e-07, + "output_cost_per_token_batches": 2e-07, "litellm_provider": "text-completion-openai", "mode": "completion" }, @@ -1396,40 +1473,40 @@ "max_tokens": 8191, "max_input_tokens": 8191, "output_vector_size": 3072, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.000000, - "input_cost_per_token_batches": 0.000000065, - "output_cost_per_token_batches": 0.000000000, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 0.0, + "input_cost_per_token_batches": 6.5e-08, + "output_cost_per_token_batches": 0.0, "litellm_provider": "openai", "mode": "embedding" }, "text-embedding-3-small": { "max_tokens": 8191, "max_input_tokens": 8191, - "output_vector_size": 1536, - "input_cost_per_token": 0.00000002, - "output_cost_per_token": 0.000000, - "input_cost_per_token_batches": 0.000000010, - "output_cost_per_token_batches": 0.000000000, + "output_vector_size": 1536, + "input_cost_per_token": 2e-08, + "output_cost_per_token": 0.0, + "input_cost_per_token_batches": 1e-08, + "output_cost_per_token_batches": 0.0, "litellm_provider": "openai", "mode": "embedding" }, "text-embedding-ada-002": { "max_tokens": 8191, "max_input_tokens": 8191, - "output_vector_size": 1536, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, + "output_vector_size": 1536, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "openai", "mode": "embedding" }, "text-embedding-ada-002-v2": { "max_tokens": 8191, "max_input_tokens": 8191, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, - "input_cost_per_token_batches": 0.000000050, - "output_cost_per_token_batches": 0.000000000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, + "input_cost_per_token_batches": 5e-08, + "output_cost_per_token_batches": 0.0, "litellm_provider": "openai", "mode": "embedding" }, @@ -1437,8 +1514,8 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 0, - "input_cost_per_token": 0.000000, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "openai", "mode": "moderation" }, @@ -1446,8 +1523,8 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 0, - "input_cost_per_token": 0.000000, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "openai", "mode": "moderation" }, @@ -1455,207 +1532,258 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 0, - "input_cost_per_token": 0.000000, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "openai", "mode": "moderation" }, "256-x-256/dall-e-2": { "mode": "image_generation", - "input_cost_per_pixel": 0.00000024414, + "input_cost_per_pixel": 2.4414e-07, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "512-x-512/dall-e-2": { "mode": "image_generation", - "input_cost_per_pixel": 0.0000000686, + "input_cost_per_pixel": 6.86e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "1024-x-1024/dall-e-2": { "mode": "image_generation", - "input_cost_per_pixel": 0.000000019, + "input_cost_per_pixel": 1.9e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "hd/1024-x-1792/dall-e-3": { "mode": "image_generation", - "input_cost_per_pixel": 0.00000006539, + "input_cost_per_pixel": 6.539e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "hd/1792-x-1024/dall-e-3": { "mode": "image_generation", - "input_cost_per_pixel": 0.00000006539, + "input_cost_per_pixel": 6.539e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "hd/1024-x-1024/dall-e-3": { "mode": "image_generation", - "input_cost_per_pixel": 0.00000007629, + "input_cost_per_pixel": 7.629e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "standard/1024-x-1792/dall-e-3": { "mode": "image_generation", - "input_cost_per_pixel": 0.00000004359, + "input_cost_per_pixel": 4.359e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "standard/1792-x-1024/dall-e-3": { "mode": "image_generation", - "input_cost_per_pixel": 0.00000004359, + "input_cost_per_pixel": 4.359e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "standard/1024-x-1024/dall-e-3": { "mode": "image_generation", - "input_cost_per_pixel": 0.0000000381469, + "input_cost_per_pixel": 3.81469e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 4.0054321e-8, + "input_cost_per_pixel": 4.0054321e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "low/1024-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.0490417e-8, + "input_cost_per_pixel": 1.0490417e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "medium/1024-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 4.0054321e-8, + "input_cost_per_pixel": 4.0054321e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "high/1024-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.59263611e-7, + "input_cost_per_pixel": 1.59263611e-07, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "low/1024-x-1536/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.0172526e-8, + "input_cost_per_pixel": 1.0172526e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "medium/1024-x-1536/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 4.0054321e-8, + "input_cost_per_pixel": 4.0054321e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "high/1024-x-1536/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.58945719e-7, + "input_cost_per_pixel": 1.58945719e-07, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "low/1536-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.0172526e-8, + "input_cost_per_pixel": 1.0172526e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "medium/1536-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 4.0054321e-8, + "input_cost_per_pixel": 4.0054321e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "high/1536-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.58945719e-7, + "input_cost_per_pixel": 1.58945719e-07, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "gpt-4o-transcribe": { "mode": "audio_transcription", "max_input_tokens": 16000, "max_output_tokens": 2000, - "input_cost_per_token": 0.0000025, - "input_cost_per_audio_token": 0.000006, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2.5e-06, + "input_cost_per_audio_token": 6e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "openai", - "supported_endpoints": ["/v1/audio/transcriptions"] - }, + "supported_endpoints": [ + "/v1/audio/transcriptions" + ] + }, "gpt-4o-mini-transcribe": { "mode": "audio_transcription", "max_input_tokens": 16000, "max_output_tokens": 2000, - "input_cost_per_token": 0.00000125, - "input_cost_per_audio_token": 0.000003, - "output_cost_per_token": 0.000005, + "input_cost_per_token": 1.25e-06, + "input_cost_per_audio_token": 3e-06, + "output_cost_per_token": 5e-06, "litellm_provider": "openai", - "supported_endpoints": ["/v1/audio/transcriptions"] - }, + "supported_endpoints": [ + "/v1/audio/transcriptions" + ] + }, "whisper-1": { "mode": "audio_transcription", "input_cost_per_second": 0.0001, - "output_cost_per_second": 0.0001, + "output_cost_per_second": 0.0001, "litellm_provider": "openai", - "supported_endpoints": ["/v1/audio/transcriptions"] - }, + "supported_endpoints": [ + "/v1/audio/transcriptions" + ] + }, "tts-1": { - "mode": "audio_speech", - "input_cost_per_character": 0.000015, + "mode": "audio_speech", + "input_cost_per_character": 1.5e-05, "litellm_provider": "openai", - "supported_endpoints": ["/v1/audio/speech"] + "supported_endpoints": [ + "/v1/audio/speech" + ] }, "tts-1-hd": { - "mode": "audio_speech", - "input_cost_per_character": 0.000030, + "mode": "audio_speech", + "input_cost_per_character": 3e-05, "litellm_provider": "openai", - "supported_endpoints": ["/v1/audio/speech"] + "supported_endpoints": [ + "/v1/audio/speech" + ] }, "gpt-4o-mini-tts": { - "mode": "audio_speech", - "input_cost_per_token": 2.5e-6, - "output_cost_per_token": 10e-6, - "output_cost_per_audio_token": 12e-6, + "mode": "audio_speech", + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_audio_token": 1.2e-05, "output_cost_per_second": 0.00025, "litellm_provider": "openai", - "supported_modalities": ["text", "audio"], - "supported_output_modalities": ["audio"], - "supported_endpoints": ["/v1/audio/speech"] + "supported_modalities": [ + "text", + "audio" + ], + "supported_output_modalities": [ + "audio" + ], + "supported_endpoints": [ + "/v1/audio/speech" + ] }, "azure/gpt-4o-mini-tts": { - "mode": "audio_speech", - "input_cost_per_token": 2.5e-6, - "output_cost_per_token": 10e-6, - "output_cost_per_audio_token": 12e-6, + "mode": "audio_speech", + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_audio_token": 1.2e-05, "output_cost_per_second": 0.00025, "litellm_provider": "azure", - "supported_modalities": ["text", "audio"], - "supported_output_modalities": ["audio"], - "supported_endpoints": ["/v1/audio/speech"] + "supported_modalities": [ + "text", + "audio" + ], + "supported_output_modalities": [ + "audio" + ], + "supported_endpoints": [ + "/v1/audio/speech" + ] }, "azure/computer-use-preview": { "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 12e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.2e-05, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -1669,15 +1797,23 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "input_cost_per_audio_token": 0.00004, - "output_cost_per_token": 0.00001, - "output_cost_per_audio_token": 0.00008, + "input_cost_per_token": 2.5e-06, + "input_cost_per_audio_token": 4e-05, + "output_cost_per_token": 1e-05, + "output_cost_per_audio_token": 8e-05, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions"], - "supported_modalities": ["text", "audio"], - "supported_output_modalities": ["text", "audio"], + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supported_modalities": [ + "text", + "audio" + ], + "supported_output_modalities": [ + "text", + "audio" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": false, @@ -1692,15 +1828,23 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "input_cost_per_audio_token": 0.00004, - "output_cost_per_token": 0.00001, - "output_cost_per_audio_token": 0.00008, + "input_cost_per_token": 2.5e-06, + "input_cost_per_audio_token": 4e-05, + "output_cost_per_token": 1e-05, + "output_cost_per_audio_token": 8e-05, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions"], - "supported_modalities": ["text", "audio"], - "supported_output_modalities": ["text", "audio"], + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supported_modalities": [ + "text", + "audio" + ], + "supported_output_modalities": [ + "text", + "audio" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": false, @@ -1715,16 +1859,25 @@ "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 2e-6, - "output_cost_per_token": 8e-6, - "input_cost_per_token_batches": 1e-6, - "output_cost_per_token_batches": 4e-6, - "cache_read_input_token_cost": 0.5e-6, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, + "input_cost_per_token_batches": 1e-06, + "output_cost_per_token_batches": 4e-06, + "cache_read_input_token_cost": 5e-07, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -1735,25 +1888,34 @@ "supports_native_streaming": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 30e-3, - "search_context_size_medium": 35e-3, - "search_context_size_high": 50e-3 + "search_context_size_low": 0.03, + "search_context_size_medium": 0.035, + "search_context_size_high": 0.05 } }, "azure/gpt-4.1-2025-04-14": { "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 2e-6, - "output_cost_per_token": 8e-6, - "input_cost_per_token_batches": 1e-6, - "output_cost_per_token_batches": 4e-6, - "cache_read_input_token_cost": 0.5e-6, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, + "input_cost_per_token_batches": 1e-06, + "output_cost_per_token_batches": 4e-06, + "cache_read_input_token_cost": 5e-07, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -1764,25 +1926,34 @@ "supports_native_streaming": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 30e-3, - "search_context_size_medium": 35e-3, - "search_context_size_high": 50e-3 + "search_context_size_low": 0.03, + "search_context_size_medium": 0.035, + "search_context_size_high": 0.05 } }, "azure/gpt-4.1-mini": { "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 0.4e-6, - "output_cost_per_token": 1.6e-6, - "input_cost_per_token_batches": 0.2e-6, - "output_cost_per_token_batches": 0.8e-6, - "cache_read_input_token_cost": 0.1e-6, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 1.6e-06, + "input_cost_per_token_batches": 2e-07, + "output_cost_per_token_batches": 8e-07, + "cache_read_input_token_cost": 1e-07, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -1793,25 +1964,34 @@ "supports_native_streaming": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 25e-3, - "search_context_size_medium": 27.5e-3, - "search_context_size_high": 30e-3 + "search_context_size_low": 0.025, + "search_context_size_medium": 0.0275, + "search_context_size_high": 0.03 } }, "azure/gpt-4.1-mini-2025-04-14": { "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 0.4e-6, - "output_cost_per_token": 1.6e-6, - "input_cost_per_token_batches": 0.2e-6, - "output_cost_per_token_batches": 0.8e-6, - "cache_read_input_token_cost": 0.1e-6, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 1.6e-06, + "input_cost_per_token_batches": 2e-07, + "output_cost_per_token_batches": 8e-07, + "cache_read_input_token_cost": 1e-07, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -1822,25 +2002,34 @@ "supports_native_streaming": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 25e-3, - "search_context_size_medium": 27.5e-3, - "search_context_size_high": 30e-3 + "search_context_size_low": 0.025, + "search_context_size_medium": 0.0275, + "search_context_size_high": 0.03 } }, "azure/gpt-4.1-nano": { "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 0.1e-6, - "output_cost_per_token": 0.4e-6, - "input_cost_per_token_batches": 0.05e-6, - "output_cost_per_token_batches": 0.2e-6, - "cache_read_input_token_cost": 0.025e-6, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, + "input_cost_per_token_batches": 5e-08, + "output_cost_per_token_batches": 2e-07, + "cache_read_input_token_cost": 2.5e-08, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -1854,16 +2043,25 @@ "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 0.1e-6, - "output_cost_per_token": 0.4e-6, - "input_cost_per_token_batches": 0.05e-6, - "output_cost_per_token_batches": 0.2e-6, - "cache_read_input_token_cost": 0.025e-6, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, + "input_cost_per_token_batches": 5e-08, + "output_cost_per_token_batches": 2e-07, + "cache_read_input_token_cost": 2.5e-08, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -1877,14 +2075,23 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 1e-5, - "output_cost_per_token": 4e-5, - "cache_read_input_token_cost": 2.5e-6, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 4e-05, + "cache_read_input_token_cost": 2.5e-06, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": false, "supports_vision": true, @@ -1897,14 +2104,23 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 1e-5, - "output_cost_per_token": 4e-5, - "cache_read_input_token_cost": 2.5e-6, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 4e-05, + "cache_read_input_token_cost": 2.5e-06, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": false, "supports_vision": true, @@ -1917,14 +2133,23 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 1.1e-6, - "output_cost_per_token": 4.4e-6, - "cache_read_input_token_cost": 2.75e-7, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 2.75e-07, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": false, "supports_vision": true, @@ -1937,12 +2162,12 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000006, - "input_cost_per_audio_token": 0.00001, - "cache_read_input_token_cost": 0.0000003, - "cache_creation_input_audio_token_cost": 0.0000003, - "output_cost_per_token": 0.0000024, - "output_cost_per_audio_token": 0.00002, + "input_cost_per_token": 6e-07, + "input_cost_per_audio_token": 1e-05, + "cache_read_input_token_cost": 3e-07, + "cache_creation_input_audio_token_cost": 3e-07, + "output_cost_per_token": 2.4e-06, + "output_cost_per_audio_token": 2e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -1956,12 +2181,12 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000066, - "input_cost_per_audio_token": 0.000011, - "cache_read_input_token_cost": 0.00000033, - "cache_creation_input_audio_token_cost": 0.00000033, - "output_cost_per_token": 0.00000264, - "output_cost_per_audio_token": 0.000022, + "input_cost_per_token": 6.6e-07, + "input_cost_per_audio_token": 1.1e-05, + "cache_read_input_token_cost": 3.3e-07, + "cache_creation_input_audio_token_cost": 3.3e-07, + "output_cost_per_token": 2.64e-06, + "output_cost_per_audio_token": 2.2e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -1975,12 +2200,12 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000066, - "input_cost_per_audio_token": 0.000011, - "cache_read_input_token_cost": 0.00000033, - "cache_creation_input_audio_token_cost": 0.00000033, - "output_cost_per_token": 0.00000264, - "output_cost_per_audio_token": 0.000022, + "input_cost_per_token": 6.6e-07, + "input_cost_per_audio_token": 1.1e-05, + "cache_read_input_token_cost": 3.3e-07, + "cache_creation_input_audio_token_cost": 3.3e-07, + "output_cost_per_token": 2.64e-06, + "output_cost_per_audio_token": 2.2e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -1994,15 +2219,21 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, - "input_cost_per_audio_token": 0.00004, - "cache_read_input_token_cost": 0.0000025, - "output_cost_per_token": 0.00002, - "output_cost_per_audio_token": 0.00008, + "input_cost_per_token": 5e-06, + "input_cost_per_audio_token": 4e-05, + "cache_read_input_token_cost": 2.5e-06, + "output_cost_per_token": 2e-05, + "output_cost_per_audio_token": 8e-05, "litellm_provider": "azure", "mode": "chat", - "supported_modalities": ["text", "audio"], - "supported_output_modalities": ["text", "audio"], + "supported_modalities": [ + "text", + "audio" + ], + "supported_output_modalities": [ + "text", + "audio" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_audio_input": true, @@ -2014,16 +2245,22 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 5.5e-6, - "input_cost_per_audio_token": 44e-6, - "cache_read_input_token_cost": 2.75e-6, - "cache_read_input_audio_token_cost": 2.5e-6, - "output_cost_per_token": 22e-6, - "output_cost_per_audio_token": 80e-6, + "input_cost_per_token": 5.5e-06, + "input_cost_per_audio_token": 4.4e-05, + "cache_read_input_token_cost": 2.75e-06, + "cache_read_input_audio_token_cost": 2.5e-06, + "output_cost_per_token": 2.2e-05, + "output_cost_per_audio_token": 8e-05, "litellm_provider": "azure", "mode": "chat", - "supported_modalities": ["text", "audio"], - "supported_output_modalities": ["text", "audio"], + "supported_modalities": [ + "text", + "audio" + ], + "supported_output_modalities": [ + "text", + "audio" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_audio_input": true, @@ -2035,16 +2272,22 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 5.5e-6, - "input_cost_per_audio_token": 44e-6, - "cache_read_input_token_cost": 2.75e-6, - "cache_read_input_audio_token_cost": 2.5e-6, - "output_cost_per_token": 22e-6, - "output_cost_per_audio_token": 80e-6, + "input_cost_per_token": 5.5e-06, + "input_cost_per_audio_token": 4.4e-05, + "cache_read_input_token_cost": 2.75e-06, + "cache_read_input_audio_token_cost": 2.5e-06, + "output_cost_per_token": 2.2e-05, + "output_cost_per_audio_token": 8e-05, "litellm_provider": "azure", "mode": "chat", - "supported_modalities": ["text", "audio"], - "supported_output_modalities": ["text", "audio"], + "supported_modalities": [ + "text", + "audio" + ], + "supported_output_modalities": [ + "text", + "audio" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_audio_input": true, @@ -2056,11 +2299,11 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, + "input_cost_per_token": 5e-06, "input_cost_per_audio_token": 0.0001, - "cache_read_input_token_cost": 0.0000025, - "cache_creation_input_audio_token_cost": 0.00002, - "output_cost_per_token": 0.00002, + "cache_read_input_token_cost": 2.5e-06, + "cache_creation_input_audio_token_cost": 2e-05, + "output_cost_per_token": 2e-05, "output_cost_per_audio_token": 0.0002, "litellm_provider": "azure", "mode": "chat", @@ -2075,11 +2318,11 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000055, + "input_cost_per_token": 5.5e-06, "input_cost_per_audio_token": 0.00011, - "cache_read_input_token_cost": 0.00000275, - "cache_creation_input_audio_token_cost": 0.000022, - "output_cost_per_token": 0.000022, + "cache_read_input_token_cost": 2.75e-06, + "cache_creation_input_audio_token_cost": 2.2e-05, + "output_cost_per_token": 2.2e-05, "output_cost_per_audio_token": 0.00022, "litellm_provider": "azure", "mode": "chat", @@ -2094,11 +2337,11 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000055, + "input_cost_per_token": 5.5e-06, "input_cost_per_audio_token": 0.00011, - "cache_read_input_token_cost": 0.00000275, - "cache_creation_input_audio_token_cost": 0.000022, - "output_cost_per_token": 0.000022, + "cache_read_input_token_cost": 2.75e-06, + "cache_creation_input_audio_token_cost": 2.2e-05, + "output_cost_per_token": 2.2e-05, "output_cost_per_audio_token": 0.00022, "litellm_provider": "azure", "mode": "chat", @@ -2113,9 +2356,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 1.1e-6, - "output_cost_per_token": 4.4e-6, - "cache_read_input_token_cost": 2.75e-7, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 2.75e-07, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2130,9 +2373,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.0000044, - "cache_read_input_token_cost": 0.00000055, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 5.5e-07, "litellm_provider": "azure", "mode": "chat", "supports_reasoning": true, @@ -2144,11 +2387,11 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.00000121, - "input_cost_per_token_batches": 0.000000605, - "output_cost_per_token": 0.00000484, - "output_cost_per_token_batches": 0.00000242, - "cache_read_input_token_cost": 0.000000605, + "input_cost_per_token": 1.21e-06, + "input_cost_per_token_batches": 6.05e-07, + "output_cost_per_token": 4.84e-06, + "output_cost_per_token_batches": 2.42e-06, + "cache_read_input_token_cost": 6.05e-07, "litellm_provider": "azure", "mode": "chat", "supports_vision": false, @@ -2160,11 +2403,11 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.00000121, - "input_cost_per_token_batches": 0.000000605, - "output_cost_per_token": 0.00000484, - "output_cost_per_token_batches": 0.00000242, - "cache_read_input_token_cost": 0.000000605, + "input_cost_per_token": 1.21e-06, + "input_cost_per_token_batches": 6.05e-07, + "output_cost_per_token": 4.84e-06, + "output_cost_per_token_batches": 2.42e-06, + "cache_read_input_token_cost": 6.05e-07, "litellm_provider": "azure", "mode": "chat", "supports_vision": false, @@ -2173,28 +2416,28 @@ "supports_tool_choice": true }, "azure/tts-1": { - "mode": "audio_speech", - "input_cost_per_character": 0.000015, + "mode": "audio_speech", + "input_cost_per_character": 1.5e-05, "litellm_provider": "azure" }, "azure/tts-1-hd": { - "mode": "audio_speech", - "input_cost_per_character": 0.000030, + "mode": "audio_speech", + "input_cost_per_character": 3e-05, "litellm_provider": "azure" }, "azure/whisper-1": { "mode": "audio_transcription", - "input_cost_per_second": 0.0001, - "output_cost_per_second": 0.0001, + "input_cost_per_second": 0.0001, + "output_cost_per_second": 0.0001, "litellm_provider": "azure" }, "azure/o3-mini": { "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.0000044, - "cache_read_input_token_cost": 0.00000055, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 5.5e-07, "litellm_provider": "azure", "mode": "chat", "supports_vision": false, @@ -2207,9 +2450,9 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.00000121, - "output_cost_per_token": 0.00000484, - "cache_read_input_token_cost": 0.000000605, + "input_cost_per_token": 1.21e-06, + "output_cost_per_token": 4.84e-06, + "cache_read_input_token_cost": 6.05e-07, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2222,9 +2465,9 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 1.1e-6, - "output_cost_per_token": 4.4e-6, - "cache_read_input_token_cost": 0.55e-6, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 5.5e-07, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2237,11 +2480,11 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.00000121, - "input_cost_per_token_batches": 0.000000605, - "output_cost_per_token": 0.00000484, - "output_cost_per_token_batches": 0.00000242, - "cache_read_input_token_cost": 0.000000605, + "input_cost_per_token": 1.21e-06, + "input_cost_per_token_batches": 6.05e-07, + "output_cost_per_token": 4.84e-06, + "output_cost_per_token_batches": 2.42e-06, + "cache_read_input_token_cost": 6.05e-07, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2253,11 +2496,11 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.00000121, - "input_cost_per_token_batches": 0.000000605, - "output_cost_per_token": 0.00000484, - "output_cost_per_token_batches": 0.00000242, - "cache_read_input_token_cost": 0.000000605, + "input_cost_per_token": 1.21e-06, + "input_cost_per_token_batches": 6.05e-07, + "output_cost_per_token": 4.84e-06, + "output_cost_per_token_batches": 2.42e-06, + "cache_read_input_token_cost": 6.05e-07, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2269,9 +2512,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2285,9 +2528,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2301,9 +2544,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.0000165, - "output_cost_per_token": 0.000066, - "cache_read_input_token_cost": 0.00000825, + "input_cost_per_token": 1.65e-05, + "output_cost_per_token": 6.6e-05, + "cache_read_input_token_cost": 8.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2316,9 +2559,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.0000165, - "output_cost_per_token": 0.000066, - "cache_read_input_token_cost": 0.00000825, + "input_cost_per_token": 1.65e-05, + "output_cost_per_token": 6.6e-05, + "cache_read_input_token_cost": 8.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2331,9 +2574,9 @@ "max_tokens": 32768, "max_input_tokens": 128000, "max_output_tokens": 32768, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2346,9 +2589,9 @@ "max_tokens": 32768, "max_input_tokens": 128000, "max_output_tokens": 32768, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "azure", "mode": "chat", "supports_pdf_input": true, @@ -2362,9 +2605,9 @@ "max_tokens": 32768, "max_input_tokens": 128000, "max_output_tokens": 32768, - "input_cost_per_token": 0.0000165, - "output_cost_per_token": 0.000066, - "cache_read_input_token_cost": 0.00000825, + "input_cost_per_token": 1.65e-05, + "output_cost_per_token": 6.6e-05, + "cache_read_input_token_cost": 8.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2376,9 +2619,9 @@ "max_tokens": 32768, "max_input_tokens": 128000, "max_output_tokens": 32768, - "input_cost_per_token": 0.0000165, - "output_cost_per_token": 0.000066, - "cache_read_input_token_cost": 0.00000825, + "input_cost_per_token": 1.65e-05, + "output_cost_per_token": 6.6e-05, + "cache_read_input_token_cost": 8.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2390,11 +2633,11 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.000075, + "input_cost_per_token": 7.5e-05, "output_cost_per_token": 0.00015, - "input_cost_per_token_batches": 0.0000375, - "output_cost_per_token_batches": 0.000075, - "cache_read_input_token_cost": 0.0000375, + "input_cost_per_token_batches": 3.75e-05, + "output_cost_per_token_batches": 7.5e-05, + "cache_read_input_token_cost": 3.75e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2409,9 +2652,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2425,9 +2668,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2441,9 +2684,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2457,9 +2700,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2473,9 +2716,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000275, - "output_cost_per_token": 0.000011, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.75e-06, + "output_cost_per_token": 1.1e-05, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2489,9 +2732,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000275, - "cache_creation_input_token_cost": 0.00000138, - "output_cost_per_token": 0.000011, + "input_cost_per_token": 2.75e-06, + "cache_creation_input_token_cost": 1.38e-06, + "output_cost_per_token": 1.1e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2504,9 +2747,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000275, - "cache_creation_input_token_cost": 0.00000138, - "output_cost_per_token": 0.000011, + "input_cost_per_token": 2.75e-06, + "cache_creation_input_token_cost": 1.38e-06, + "output_cost_per_token": 1.1e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2519,8 +2762,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2533,9 +2776,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000010, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2550,9 +2793,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000275, - "output_cost_per_token": 0.000011, - "cache_read_input_token_cost": 0.000001375, + "input_cost_per_token": 2.75e-06, + "output_cost_per_token": 1.1e-05, + "cache_read_input_token_cost": 1.375e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2566,9 +2809,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000275, - "output_cost_per_token": 0.000011, - "cache_read_input_token_cost": 0.000001375, + "input_cost_per_token": 2.75e-06, + "output_cost_per_token": 1.1e-05, + "cache_read_input_token_cost": 1.375e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2582,9 +2825,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000010, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2598,8 +2841,8 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000060, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2612,9 +2855,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.000000165, - "output_cost_per_token": 0.00000066, - "cache_read_input_token_cost": 0.000000075, + "input_cost_per_token": 1.65e-07, + "output_cost_per_token": 6.6e-07, + "cache_read_input_token_cost": 7.5e-08, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2628,9 +2871,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.000000165, - "output_cost_per_token": 0.00000066, - "cache_read_input_token_cost": 0.000000075, + "input_cost_per_token": 1.65e-07, + "output_cost_per_token": 6.6e-07, + "cache_read_input_token_cost": 7.5e-08, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2644,9 +2887,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.000000165, - "output_cost_per_token": 0.00000066, - "cache_read_input_token_cost": 0.000000083, + "input_cost_per_token": 1.65e-07, + "output_cost_per_token": 6.6e-07, + "cache_read_input_token_cost": 8.3e-08, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2660,9 +2903,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.000000165, - "output_cost_per_token": 0.00000066, - "cache_read_input_token_cost": 0.000000083, + "input_cost_per_token": 1.65e-07, + "output_cost_per_token": 6.6e-07, + "cache_read_input_token_cost": 8.3e-08, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2676,8 +2919,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2689,8 +2932,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2701,8 +2944,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2713,8 +2956,8 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.00006, + "input_cost_per_token": 3e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2724,7 +2967,7 @@ "max_tokens": 4096, "max_input_tokens": 32768, "max_output_tokens": 4096, - "input_cost_per_token": 0.00006, + "input_cost_per_token": 6e-05, "output_cost_per_token": 0.00012, "litellm_provider": "azure", "mode": "chat", @@ -2734,7 +2977,7 @@ "max_tokens": 4096, "max_input_tokens": 32768, "max_output_tokens": 4096, - "input_cost_per_token": 0.00006, + "input_cost_per_token": 6e-05, "output_cost_per_token": 0.00012, "litellm_provider": "azure", "mode": "chat", @@ -2744,8 +2987,8 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.00006, + "input_cost_per_token": 3e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2755,9 +2998,9 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, - "litellm_provider": "azure", + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, + "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -2767,9 +3010,9 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, - "litellm_provider": "azure", + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, + "litellm_provider": "azure", "mode": "chat", "supports_vision": true, "supports_tool_choice": true @@ -2778,8 +3021,8 @@ "max_tokens": 4096, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000004, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 4e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2789,8 +3032,8 @@ "max_tokens": 4096, "max_input_tokens": 16384, "max_output_tokens": 4096, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2802,8 +3045,8 @@ "max_tokens": 4097, "max_input_tokens": 4097, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2815,8 +3058,8 @@ "max_tokens": 4097, "max_input_tokens": 4097, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2828,8 +3071,8 @@ "max_tokens": 4096, "max_input_tokens": 16384, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2841,8 +3084,8 @@ "max_tokens": 4096, "max_input_tokens": 16384, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2854,8 +3097,8 @@ "max_tokens": 4096, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000004, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 4e-06, "litellm_provider": "azure", "mode": "chat", "supports_tool_choice": true @@ -2864,8 +3107,8 @@ "max_tokens": 4096, "max_input_tokens": 4097, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2875,8 +3118,8 @@ "max_tokens": 4096, "max_input_tokens": 4097, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2885,32 +3128,32 @@ "azure/gpt-3.5-turbo-instruct-0914": { "max_tokens": 4097, "max_input_tokens": 4097, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "azure_text", "mode": "completion" }, "azure/gpt-35-turbo-instruct": { "max_tokens": 4097, "max_input_tokens": 4097, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "azure_text", "mode": "completion" }, "azure/gpt-35-turbo-instruct-0914": { "max_tokens": 4097, "max_input_tokens": 4097, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "azure_text", "mode": "completion" }, "azure/mistral-large-latest": { "max_tokens": 32000, "max_input_tokens": 32000, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true @@ -2918,18 +3161,18 @@ "azure/mistral-large-2402": { "max_tokens": 32000, "max_input_tokens": 32000, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true }, "azure/command-r-plus": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true @@ -2937,153 +3180,173 @@ "azure/ada": { "max_tokens": 8191, "max_input_tokens": 8191, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "azure", "mode": "embedding" }, "azure/text-embedding-ada-002": { "max_tokens": 8191, "max_input_tokens": 8191, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "azure", "mode": "embedding" }, "azure/text-embedding-3-large": { "max_tokens": 8191, "max_input_tokens": 8191, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 0.0, "litellm_provider": "azure", "mode": "embedding" }, "azure/text-embedding-3-small": { "max_tokens": 8191, "max_input_tokens": 8191, - "input_cost_per_token": 0.00000002, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 2e-08, + "output_cost_per_token": 0.0, "litellm_provider": "azure", "mode": "embedding" }, "azure/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 4.0054321e-8, + "input_cost_per_pixel": 4.0054321e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/low/1024-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.0490417e-8, + "input_cost_per_pixel": 1.0490417e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/medium/1024-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 4.0054321e-8, + "input_cost_per_pixel": 4.0054321e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/high/1024-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.59263611e-7, + "input_cost_per_pixel": 1.59263611e-07, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/low/1024-x-1536/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.0172526e-8, + "input_cost_per_pixel": 1.0172526e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/medium/1024-x-1536/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 4.0054321e-8, + "input_cost_per_pixel": 4.0054321e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/high/1024-x-1536/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.58945719e-7, + "input_cost_per_pixel": 1.58945719e-07, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/low/1536-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.0172526e-8, + "input_cost_per_pixel": 1.0172526e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/medium/1536-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 4.0054321e-8, + "input_cost_per_pixel": 4.0054321e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/high/1536-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.58945719e-7, + "input_cost_per_pixel": 1.58945719e-07, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] - }, + "supported_endpoints": [ + "/v1/images/generations" + ] + }, "azure/standard/1024-x-1024/dall-e-3": { - "input_cost_per_pixel": 0.0000000381469, + "input_cost_per_pixel": 3.81469e-08, "output_cost_per_token": 0.0, - "litellm_provider": "azure", + "litellm_provider": "azure", "mode": "image_generation" }, "azure/hd/1024-x-1024/dall-e-3": { - "input_cost_per_pixel": 0.00000007629, + "input_cost_per_pixel": 7.629e-08, "output_cost_per_token": 0.0, - "litellm_provider": "azure", + "litellm_provider": "azure", "mode": "image_generation" }, "azure/standard/1024-x-1792/dall-e-3": { - "input_cost_per_pixel": 0.00000004359, + "input_cost_per_pixel": 4.359e-08, "output_cost_per_token": 0.0, - "litellm_provider": "azure", + "litellm_provider": "azure", "mode": "image_generation" }, "azure/standard/1792-x-1024/dall-e-3": { - "input_cost_per_pixel": 0.00000004359, + "input_cost_per_pixel": 4.359e-08, "output_cost_per_token": 0.0, - "litellm_provider": "azure", + "litellm_provider": "azure", "mode": "image_generation" }, "azure/hd/1024-x-1792/dall-e-3": { - "input_cost_per_pixel": 0.00000006539, + "input_cost_per_pixel": 6.539e-08, "output_cost_per_token": 0.0, - "litellm_provider": "azure", + "litellm_provider": "azure", "mode": "image_generation" }, "azure/hd/1792-x-1024/dall-e-3": { - "input_cost_per_pixel": 0.00000006539, + "input_cost_per_pixel": 6.539e-08, "output_cost_per_token": 0.0, - "litellm_provider": "azure", + "litellm_provider": "azure", "mode": "image_generation" }, "azure/standard/1024-x-1024/dall-e-2": { "input_cost_per_pixel": 0.0, "output_cost_per_token": 0.0, - "litellm_provider": "azure", + "litellm_provider": "azure", "mode": "image_generation" }, "azure_ai/deepseek-r1": { "max_tokens": 8192, "max_input_tokens": 128000, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000135, - "output_cost_per_token": 0.0000054, + "input_cost_per_token": 1.35e-06, + "output_cost_per_token": 5.4e-06, "litellm_provider": "azure_ai", "mode": "chat", "supports_tool_choice": true, @@ -3094,8 +3357,8 @@ "max_tokens": 8192, "max_input_tokens": 128000, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000114, - "output_cost_per_token": 0.00000456, + "input_cost_per_token": 1.14e-06, + "output_cost_per_token": 4.56e-06, "litellm_provider": "azure_ai", "mode": "chat", "supports_tool_choice": true, @@ -3105,8 +3368,8 @@ "max_tokens": 8192, "max_input_tokens": 128000, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000114, - "output_cost_per_token": 0.00000456, + "input_cost_per_token": 1.14e-06, + "output_cost_per_token": 4.56e-06, "litellm_provider": "azure_ai", "mode": "chat", "supports_function_calling": true, @@ -3117,8 +3380,8 @@ "max_tokens": 4096, "max_input_tokens": 70000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000007, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_tool_choice": true @@ -3127,8 +3390,8 @@ "max_tokens": 4096, "max_input_tokens": 131072, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_function_calling": true, @@ -3138,8 +3401,8 @@ "max_tokens": 8191, "max_input_tokens": 131072, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000004, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 2e-06, "litellm_provider": "azure_ai", "mode": "chat", "supports_function_calling": true, @@ -3150,8 +3413,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000004, - "output_cost_per_token": 0.000012, + "input_cost_per_token": 4e-06, + "output_cost_per_token": 1.2e-05, "litellm_provider": "azure_ai", "mode": "chat", "supports_function_calling": true, @@ -3161,8 +3424,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "azure_ai", "supports_function_calling": true, "mode": "chat", @@ -3172,8 +3435,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "azure_ai", "mode": "chat", "supports_function_calling": true, @@ -3184,8 +3447,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "azure_ai", "supports_function_calling": true, "mode": "chat", @@ -3196,8 +3459,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "azure_ai", "supports_function_calling": true, "mode": "chat", @@ -3208,20 +3471,20 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000004, - "output_cost_per_token": 0.00000004, + "input_cost_per_token": 4e-08, + "output_cost_per_token": 4e-08, "litellm_provider": "azure_ai", "supports_function_calling": true, "mode": "chat", "source": "https://azuremarketplace.microsoft.com/en/marketplace/apps/000-000.ministral-3b-2410-offer?tab=Overview", "supports_tool_choice": true - }, + }, "azure_ai/Llama-3.2-11B-Vision-Instruct": { "max_tokens": 2048, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000037, - "output_cost_per_token": 0.00000037, + "input_cost_per_token": 3.7e-07, + "output_cost_per_token": 3.7e-07, "litellm_provider": "azure_ai", "supports_function_calling": true, "supports_vision": true, @@ -3233,8 +3496,8 @@ "max_tokens": 2048, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000071, - "output_cost_per_token": 0.00000071, + "input_cost_per_token": 7.1e-07, + "output_cost_per_token": 7.1e-07, "litellm_provider": "azure_ai", "supports_function_calling": true, "mode": "chat", @@ -3245,8 +3508,8 @@ "max_tokens": 16384, "max_input_tokens": 10000000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.00000078, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 7.8e-07, "litellm_provider": "azure_ai", "supports_function_calling": true, "supports_vision": true, @@ -3258,8 +3521,8 @@ "max_tokens": 16384, "max_input_tokens": 1000000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000141, - "output_cost_per_token": 0.00000035, + "input_cost_per_token": 1.41e-06, + "output_cost_per_token": 3.5e-07, "litellm_provider": "azure_ai", "supports_function_calling": true, "supports_vision": true, @@ -3271,8 +3534,8 @@ "max_tokens": 2048, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000204, - "output_cost_per_token": 0.00000204, + "input_cost_per_token": 2.04e-06, + "output_cost_per_token": 2.04e-06, "litellm_provider": "azure_ai", "supports_function_calling": true, "supports_vision": true, @@ -3284,8 +3547,8 @@ "max_tokens": 2048, "max_input_tokens": 8192, "max_output_tokens": 2048, - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.00000037, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 3.7e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_tool_choice": true @@ -3294,41 +3557,41 @@ "max_tokens": 2048, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.00000061, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 6.1e-07, "litellm_provider": "azure_ai", "mode": "chat", - "source":"https://azuremarketplace.microsoft.com/en-us/marketplace/apps/metagenai.meta-llama-3-1-8b-instruct-offer?tab=PlansAndPrice", + "source": "https://azuremarketplace.microsoft.com/en-us/marketplace/apps/metagenai.meta-llama-3-1-8b-instruct-offer?tab=PlansAndPrice", "supports_tool_choice": true }, "azure_ai/Meta-Llama-3.1-70B-Instruct": { "max_tokens": 2048, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000268, - "output_cost_per_token": 0.00000354, + "input_cost_per_token": 2.68e-06, + "output_cost_per_token": 3.54e-06, "litellm_provider": "azure_ai", "mode": "chat", - "source":"https://azuremarketplace.microsoft.com/en-us/marketplace/apps/metagenai.meta-llama-3-1-70b-instruct-offer?tab=PlansAndPrice", + "source": "https://azuremarketplace.microsoft.com/en-us/marketplace/apps/metagenai.meta-llama-3-1-70b-instruct-offer?tab=PlansAndPrice", "supports_tool_choice": true }, "azure_ai/Meta-Llama-3.1-405B-Instruct": { "max_tokens": 2048, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000533, - "output_cost_per_token": 0.000016, + "input_cost_per_token": 5.33e-06, + "output_cost_per_token": 1.6e-05, "litellm_provider": "azure_ai", "mode": "chat", - "source":"https://azuremarketplace.microsoft.com/en-us/marketplace/apps/metagenai.meta-llama-3-1-405b-instruct-offer?tab=PlansAndPrice", + "source": "https://azuremarketplace.microsoft.com/en-us/marketplace/apps/metagenai.meta-llama-3-1-405b-instruct-offer?tab=PlansAndPrice", "supports_tool_choice": true }, "azure_ai/Phi-4-mini-instruct": { "max_tokens": 4096, "max_input_tokens": 131072, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000075, - "output_cost_per_token": 0.0000003, + "input_cost_per_token": 7.5e-08, + "output_cost_per_token": 3e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_function_calling": true, @@ -3338,9 +3601,9 @@ "max_tokens": 4096, "max_input_tokens": 131072, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000008, - "input_cost_per_audio_token": 0.000004, - "output_cost_per_token": 0.00000032, + "input_cost_per_token": 8e-08, + "input_cost_per_audio_token": 4e-06, + "output_cost_per_token": 3.2e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_audio_input": true, @@ -3352,8 +3615,8 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3365,8 +3628,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000052, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 5.2e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3377,8 +3640,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000052, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 5.2e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": true, @@ -3389,8 +3652,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000016, - "output_cost_per_token": 0.00000064, + "input_cost_per_token": 1.6e-07, + "output_cost_per_token": 6.4e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3401,8 +3664,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000052, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 5.2e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3413,8 +3676,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000052, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 5.2e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3425,8 +3688,8 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3437,8 +3700,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3449,8 +3712,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000017, - "output_cost_per_token": 0.00000068, + "input_cost_per_token": 1.7e-07, + "output_cost_per_token": 6.8e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3461,8 +3724,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000017, - "output_cost_per_token": 0.00000068, + "input_cost_per_token": 1.7e-07, + "output_cost_per_token": 6.8e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3495,43 +3758,48 @@ "max_tokens": 512, "max_input_tokens": 512, "output_vector_size": 1024, - "input_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0.0, "litellm_provider": "azure_ai", "mode": "embedding", "supports_embedding_image_input": true, - "source":"https://azuremarketplace.microsoft.com/en-us/marketplace/apps/cohere.cohere-embed-v3-english-offer?tab=PlansAndPrice" + "source": "https://azuremarketplace.microsoft.com/en-us/marketplace/apps/cohere.cohere-embed-v3-english-offer?tab=PlansAndPrice" }, "azure_ai/Cohere-embed-v3-multilingual": { "max_tokens": 512, "max_input_tokens": 512, "output_vector_size": 1024, - "input_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0.0, "litellm_provider": "azure_ai", "mode": "embedding", "supports_embedding_image_input": true, - "source":"https://azuremarketplace.microsoft.com/en-us/marketplace/apps/cohere.cohere-embed-v3-english-offer?tab=PlansAndPrice" + "source": "https://azuremarketplace.microsoft.com/en-us/marketplace/apps/cohere.cohere-embed-v3-english-offer?tab=PlansAndPrice" }, "azure_ai/embed-v-4-0": { "max_tokens": 128000, "max_input_tokens": 128000, "output_vector_size": 3072, - "input_cost_per_token": 0.00000012, + "input_cost_per_token": 1.2e-07, "output_cost_per_token": 0.0, "litellm_provider": "azure_ai", "mode": "embedding", "supports_embedding_image_input": true, - "supported_endpoints": ["/v1/embeddings"], - "supported_modalities": ["text", "image"], - "source":"https://azuremarketplace.microsoft.com/pt-br/marketplace/apps/cohere.cohere-embed-4-offer?tab=PlansAndPrice" + "supported_endpoints": [ + "/v1/embeddings" + ], + "supported_modalities": [ + "text", + "image" + ], + "source": "https://azuremarketplace.microsoft.com/pt-br/marketplace/apps/cohere.cohere-embed-4-offer?tab=PlansAndPrice" }, "babbage-002": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000004, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "text-completion-openai", "mode": "completion" }, @@ -3539,17 +3807,17 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 4096, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "text-completion-openai", "mode": "completion" - }, + }, "gpt-3.5-turbo-instruct": { "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "text-completion-openai", "mode": "completion" }, @@ -3557,18 +3825,17 @@ "max_tokens": 4097, "max_input_tokens": 8192, "max_output_tokens": 4097, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "text-completion-openai", "mode": "completion" - }, "claude-instant-1": { "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000163, - "output_cost_per_token": 0.00000551, + "input_cost_per_token": 1.63e-06, + "output_cost_per_token": 5.51e-06, "litellm_provider": "anthropic", "mode": "chat" }, @@ -3576,8 +3843,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 2.5e-07, "litellm_provider": "mistral", "mode": "chat", "supports_assistant_prefill": true, @@ -3587,8 +3854,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000003, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 3e-07, "litellm_provider": "mistral", "supports_function_calling": true, "mode": "chat", @@ -3599,8 +3866,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000003, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 3e-07, "litellm_provider": "mistral", "supports_function_calling": true, "mode": "chat", @@ -3611,8 +3878,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000027, - "output_cost_per_token": 0.0000081, + "input_cost_per_token": 2.7e-06, + "output_cost_per_token": 8.1e-06, "litellm_provider": "mistral", "mode": "chat", "supports_assistant_prefill": true, @@ -3622,8 +3889,8 @@ "max_tokens": 8191, "max_input_tokens": 131072, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000004, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 2e-06, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3634,8 +3901,8 @@ "max_tokens": 8191, "max_input_tokens": 131072, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000004, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 2e-06, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3646,8 +3913,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000027, - "output_cost_per_token": 0.0000081, + "input_cost_per_token": 2.7e-06, + "output_cost_per_token": 8.1e-06, "litellm_provider": "mistral", "mode": "chat", "supports_assistant_prefill": true, @@ -3657,8 +3924,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3669,8 +3936,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3681,8 +3948,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000004, - "output_cost_per_token": 0.000012, + "input_cost_per_token": 4e-06, + "output_cost_per_token": 1.2e-05, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3693,8 +3960,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000009, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 9e-06, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3705,8 +3972,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3718,8 +3985,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3731,8 +3998,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3744,8 +4011,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 2.5e-07, "litellm_provider": "mistral", "mode": "chat", "supports_assistant_prefill": true, @@ -3755,8 +4022,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000007, - "output_cost_per_token": 0.0000007, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3767,8 +4034,8 @@ "max_tokens": 8191, "max_input_tokens": 65336, "max_output_tokens": 8191, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3779,8 +4046,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "mistral", "mode": "chat", "supports_assistant_prefill": true, @@ -3790,8 +4057,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "mistral", "mode": "chat", "supports_assistant_prefill": true, @@ -3801,8 +4068,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000003, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 3e-07, "litellm_provider": "mistral", "mode": "chat", "source": "https://mistral.ai/technology/", @@ -3813,8 +4080,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000003, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 3e-07, "litellm_provider": "mistral", "mode": "chat", "source": "https://mistral.ai/technology/", @@ -3825,8 +4092,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 2.5e-07, "litellm_provider": "mistral", "mode": "chat", "source": "https://mistral.ai/technology/", @@ -3837,8 +4104,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 2.5e-07, "litellm_provider": "mistral", "mode": "chat", "source": "https://mistral.ai/technology/", @@ -3849,8 +4116,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000003, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 3e-07, "litellm_provider": "mistral", "mode": "chat", "source": "https://mistral.ai/news/devstral", @@ -3861,7 +4128,7 @@ "mistral/mistral-embed": { "max_tokens": 8192, "max_input_tokens": 8192, - "input_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, "litellm_provider": "mistral", "mode": "embedding" }, @@ -3869,12 +4136,12 @@ "max_tokens": 8192, "max_input_tokens": 65536, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000055, - "input_cost_per_token_cache_hit": 0.00000014, - "output_cost_per_token": 0.00000219, + "input_cost_per_token": 5.5e-07, + "input_cost_per_token_cache_hit": 1.4e-07, + "output_cost_per_token": 2.19e-06, "litellm_provider": "deepseek", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_assistant_prefill": true, "supports_tool_choice": true, "supports_reasoning": true, @@ -3884,14 +4151,14 @@ "max_tokens": 8192, "max_input_tokens": 65536, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000027, - "input_cost_per_token_cache_hit": 0.00000007, - "cache_read_input_token_cost": 0.00000007, + "input_cost_per_token": 2.7e-07, + "input_cost_per_token_cache_hit": 7e-08, + "cache_read_input_token_cost": 7e-08, "cache_creation_input_token_cost": 0.0, - "output_cost_per_token": 0.0000011, + "output_cost_per_token": 1.1e-06, "litellm_provider": "deepseek", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_assistant_prefill": true, "supports_tool_choice": true, "supports_prompt_caching": true @@ -3900,8 +4167,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000000, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "codestral", "mode": "chat", "source": "https://docs.mistral.ai/capabilities/code_generation/", @@ -3912,8 +4179,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000000, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "codestral", "mode": "chat", "source": "https://docs.mistral.ai/capabilities/code_generation/", @@ -3924,8 +4191,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000000, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "text-completion-codestral", "mode": "completion", "source": "https://docs.mistral.ai/capabilities/code_generation/" @@ -3934,8 +4201,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000000, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "text-completion-codestral", "mode": "completion", "source": "https://docs.mistral.ai/capabilities/code_generation/" @@ -3944,8 +4211,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -3956,9 +4223,9 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 32768, - "input_cost_per_token": 0.000002, - "input_cost_per_image": 0.000002, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2e-06, + "input_cost_per_image": 2e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -3969,9 +4236,9 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 32768, - "input_cost_per_token": 0.000002, - "input_cost_per_image": 0.000002, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2e-06, + "input_cost_per_image": 2e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -3982,9 +4249,9 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 32768, - "input_cost_per_token": 0.000002, - "input_cost_per_image": 0.000002, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2e-06, + "input_cost_per_image": 2e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -3995,8 +4262,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4008,8 +4275,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4021,8 +4288,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000025, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 2.5e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4034,8 +4301,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000025, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 2.5e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4047,8 +4314,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4061,8 +4328,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.000004, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 4e-06, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4075,8 +4342,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.000004, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 4e-06, "litellm_provider": "xai", "mode": "chat", "supports_reasoning": true, @@ -4089,9 +4356,9 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.000005, - "input_cost_per_image": 0.000005, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 5e-06, + "input_cost_per_image": 5e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4102,8 +4369,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4113,8 +4380,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4124,8 +4391,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4135,12 +4402,12 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000014, - "input_cost_per_token_cache_hit": 0.000000014, - "output_cost_per_token": 0.00000028, + "input_cost_per_token": 1.4e-07, + "input_cost_per_token_cache_hit": 1.4e-08, + "output_cost_per_token": 2.8e-07, "litellm_provider": "deepseek", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_assistant_prefill": true, "supports_tool_choice": true, "supports_prompt_caching": true @@ -4194,8 +4461,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000080, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 8e-07, "litellm_provider": "groq", "mode": "chat", "supports_function_calling": true, @@ -4332,8 +4599,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000059, - "output_cost_per_token": 0.00000079, + "input_cost_per_token": 5.9e-07, + "output_cost_per_token": 7.9e-07, "litellm_provider": "groq", "mode": "chat", "supports_function_calling": true, @@ -4480,8 +4747,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 1e-07, "litellm_provider": "cerebras", "mode": "chat", "supports_function_calling": true, @@ -4491,8 +4758,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "cerebras", "mode": "chat", "supports_function_calling": true, @@ -4502,8 +4769,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.00000085, - "output_cost_per_token": 0.0000012, + "input_cost_per_token": 8.5e-07, + "output_cost_per_token": 1.2e-06, "litellm_provider": "cerebras", "mode": "chat", "supports_function_calling": true, @@ -4513,8 +4780,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 1e-07, "litellm_provider": "friendliai", "mode": "chat", "supports_function_calling": true, @@ -4527,8 +4794,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "friendliai", "mode": "chat", "supports_function_calling": true, @@ -4541,8 +4808,8 @@ "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000000163, - "output_cost_per_token": 0.000000551, + "input_cost_per_token": 1.63e-07, + "output_cost_per_token": 5.51e-07, "litellm_provider": "anthropic", "mode": "chat", "supports_tool_choice": true @@ -4551,8 +4818,8 @@ "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "anthropic", "mode": "chat" }, @@ -4560,8 +4827,8 @@ "max_tokens": 8191, "max_input_tokens": 200000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "anthropic", "mode": "chat", "supports_tool_choice": true @@ -4570,10 +4837,10 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, - "cache_creation_input_token_cost": 0.0000003, - "cache_read_input_token_cost": 0.00000003, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, + "cache_creation_input_token_cost": 3e-07, + "cache_read_input_token_cost": 3e-08, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4589,14 +4856,14 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.000004, - "cache_creation_input_token_cost": 0.000001, - "cache_read_input_token_cost": 0.00000008, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 4e-06, + "cache_creation_input_token_cost": 1e-06, + "cache_read_input_token_cost": 8e-08, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, "litellm_provider": "anthropic", "mode": "chat", @@ -4615,14 +4882,14 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000005, - "cache_creation_input_token_cost": 0.00000125, - "cache_read_input_token_cost": 0.0000001, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 5e-06, + "cache_creation_input_token_cost": 1.25e-06, + "cache_read_input_token_cost": 1e-07, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, "litellm_provider": "anthropic", "mode": "chat", @@ -4641,10 +4908,10 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000075, - "cache_creation_input_token_cost": 0.00001875, - "cache_read_input_token_cost": 0.0000015, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, + "cache_creation_input_token_cost": 1.875e-05, + "cache_read_input_token_cost": 1.5e-06, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4660,10 +4927,10 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000075, - "cache_creation_input_token_cost": 0.00001875, - "cache_read_input_token_cost": 0.0000015, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, + "cache_creation_input_token_cost": 1.875e-05, + "cache_read_input_token_cost": 1.5e-06, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4679,8 +4946,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4697,14 +4964,14 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, "litellm_provider": "anthropic", "mode": "chat", @@ -4723,10 +4990,10 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4743,15 +5010,15 @@ "max_tokens": 32000, "max_input_tokens": 200000, "max_output_tokens": 32000, - "input_cost_per_token": 15e-6, - "output_cost_per_token": 75e-6, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 18.75e-6, - "cache_read_input_token_cost": 1.5e-6, + "cache_creation_input_token_cost": 1.875e-05, + "cache_read_input_token_cost": 1.5e-06, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4769,15 +5036,15 @@ "max_tokens": 64000, "max_input_tokens": 200000, "max_output_tokens": 64000, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 15e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 3.75e-6, - "cache_read_input_token_cost": 0.3e-6, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4796,15 +5063,15 @@ "max_tokens": 128000, "max_input_tokens": 200000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4823,14 +5090,14 @@ "max_tokens": 128000, "max_input_tokens": 200000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, "litellm_provider": "anthropic", "mode": "chat", @@ -4851,14 +5118,14 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, "litellm_provider": "anthropic", "mode": "chat", @@ -4877,8 +5144,8 @@ "max_tokens": 2048, "max_input_tokens": 8192, "max_output_tokens": 2048, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -4887,8 +5154,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -4897,8 +5164,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -4907,10 +5174,10 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -4919,10 +5186,10 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -4931,8 +5198,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.000028, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 2.8e-05, "litellm_provider": "vertex_ai-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -4941,8 +5208,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.000028, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 2.8e-05, "litellm_provider": "vertex_ai-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -4951,10 +5218,10 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -4964,10 +5231,10 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -4977,10 +5244,10 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -4991,10 +5258,10 @@ "max_tokens": 8192, "max_input_tokens": 32000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5004,10 +5271,10 @@ "max_tokens": 8192, "max_input_tokens": 32000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5017,10 +5284,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5030,10 +5297,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -5042,10 +5309,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -5054,10 +5321,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -5066,10 +5333,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -5078,8 +5345,8 @@ "max_tokens": 64, "max_input_tokens": 2048, "max_output_tokens": 64, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -5088,8 +5355,8 @@ "max_tokens": 64, "max_input_tokens": 2048, "max_output_tokens": 64, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -5098,8 +5365,8 @@ "max_tokens": 64, "max_input_tokens": 2048, "max_output_tokens": 64, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -5108,8 +5375,8 @@ "max_tokens": 64, "max_input_tokens": 2048, "max_output_tokens": 64, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -5118,10 +5385,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5131,10 +5398,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5144,10 +5411,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5157,10 +5424,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5170,10 +5437,10 @@ "max_tokens": 8192, "max_input_tokens": 32000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5183,10 +5450,10 @@ "max_tokens": 8192, "max_input_tokens": 32000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5201,8 +5468,13 @@ "supports_function_calling": false, "source": "https://llama.developer.meta.com/docs/models", "supports_tool_choice": false, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ] }, "meta_llama/Llama-4-Maverick-17B-128E-Instruct-FP8": { "max_tokens": 128000, @@ -5213,8 +5485,13 @@ "supports_function_calling": false, "source": "https://llama.developer.meta.com/docs/models", "supports_tool_choice": false, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ] }, "meta_llama/Llama-3.3-70B-Instruct": { "max_tokens": 128000, @@ -5225,8 +5502,12 @@ "supports_function_calling": false, "source": "https://llama.developer.meta.com/docs/models", "supports_tool_choice": false, - "supported_modalities": ["text"], - "supported_output_modalities": ["text"] + "supported_modalities": [ + "text" + ], + "supported_output_modalities": [ + "text" + ] }, "meta_llama/Llama-3.3-8B-Instruct": { "max_tokens": 128000, @@ -5237,8 +5518,12 @@ "supports_function_calling": false, "source": "https://llama.developer.meta.com/docs/models", "supports_tool_choice": false, - "supported_modalities": ["text"], - "supported_output_modalities": ["text"] + "supported_modalities": [ + "text" + ], + "supported_output_modalities": [ + "text" + ] }, "gemini-pro": { "max_tokens": 8192, @@ -5246,48 +5531,51 @@ "max_output_tokens": 8192, "input_cost_per_image": 0.0025, "input_cost_per_video_per_second": 0.002, - "input_cost_per_token": 0.0000005, - "input_cost_per_character": 0.000000125, - "output_cost_per_token": 0.0000015, - "output_cost_per_character": 0.000000375, + "input_cost_per_token": 5e-07, + "input_cost_per_character": 1.25e-07, + "output_cost_per_token": 1.5e-06, + "output_cost_per_character": 3.75e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": true, + "supports_parallel_function_calling": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing", "supports_tool_choice": true }, - "gemini-1.0-pro": { + "gemini-1.0-pro": { "max_tokens": 8192, "max_input_tokens": 32760, "max_output_tokens": 8192, "input_cost_per_image": 0.0025, "input_cost_per_video_per_second": 0.002, - "input_cost_per_token": 0.0000005, - "input_cost_per_character": 0.000000125, - "output_cost_per_token": 0.0000015, - "output_cost_per_character": 0.000000375, + "input_cost_per_token": 5e-07, + "input_cost_per_character": 1.25e-07, + "output_cost_per_token": 1.5e-06, + "output_cost_per_character": 3.75e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": true, + "supports_parallel_function_calling": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#google_models", "supports_tool_choice": true }, - "gemini-1.0-pro-001": { + "gemini-1.0-pro-001": { "max_tokens": 8192, "max_input_tokens": 32760, "max_output_tokens": 8192, "input_cost_per_image": 0.0025, "input_cost_per_video_per_second": 0.002, - "input_cost_per_token": 0.0000005, - "input_cost_per_character": 0.000000125, - "output_cost_per_token": 0.0000015, - "output_cost_per_character": 0.000000375, + "input_cost_per_token": 5e-07, + "input_cost_per_character": 1.25e-07, + "output_cost_per_token": 1.5e-06, + "output_cost_per_character": 3.75e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", "deprecation_date": "2025-04-09", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-1.0-ultra": { "max_tokens": 8192, @@ -5295,15 +5583,16 @@ "max_output_tokens": 2048, "input_cost_per_image": 0.0025, "input_cost_per_video_per_second": 0.002, - "input_cost_per_token": 0.0000005, - "input_cost_per_character": 0.000000125, - "output_cost_per_token": 0.0000015, - "output_cost_per_character": 0.000000375, + "input_cost_per_token": 5e-07, + "input_cost_per_character": 1.25e-07, + "output_cost_per_token": 1.5e-06, + "output_cost_per_character": 3.75e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": true, "source": "As of Jun, 2024. There is no available doc on vertex ai pricing gemini-1.0-ultra-001. Using gemini-1.0-pro pricing. Got max_tokens info here: https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-1.0-ultra-001": { "max_tokens": 8192, @@ -5311,193 +5600,201 @@ "max_output_tokens": 2048, "input_cost_per_image": 0.0025, "input_cost_per_video_per_second": 0.002, - "input_cost_per_token": 0.0000005, - "input_cost_per_character": 0.000000125, - "output_cost_per_token": 0.0000015, - "output_cost_per_character": 0.000000375, + "input_cost_per_token": 5e-07, + "input_cost_per_character": 1.25e-07, + "output_cost_per_token": 1.5e-06, + "output_cost_per_character": 3.75e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": true, "source": "As of Jun, 2024. There is no available doc on vertex ai pricing gemini-1.0-ultra-001. Using gemini-1.0-pro pricing. Got max_tokens info here: https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, - "gemini-1.0-pro-002": { + "gemini-1.0-pro-002": { "max_tokens": 8192, "max_input_tokens": 32760, "max_output_tokens": 8192, "input_cost_per_image": 0.0025, "input_cost_per_video_per_second": 0.002, - "input_cost_per_token": 0.0000005, - "input_cost_per_character": 0.000000125, - "output_cost_per_token": 0.0000015, - "output_cost_per_character": 0.000000375, + "input_cost_per_token": 5e-07, + "input_cost_per_character": 1.25e-07, + "output_cost_per_token": 1.5e-06, + "output_cost_per_character": 3.75e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", "deprecation_date": "2025-04-09", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, - "gemini-1.5-pro": { + "gemini-1.5-pro": { "max_tokens": 8192, "max_input_tokens": 2097152, "max_output_tokens": 8192, "input_cost_per_image": 0.00032875, - "input_cost_per_audio_per_second": 0.00003125, + "input_cost_per_audio_per_second": 3.125e-05, "input_cost_per_video_per_second": 0.00032875, - "input_cost_per_token": 0.00000125, - "input_cost_per_character": 0.0000003125, - "input_cost_per_image_above_128k_tokens": 0.0006575, - "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, - "input_cost_per_audio_per_second_above_128k_tokens": 0.0000625, - "input_cost_per_token_above_128k_tokens": 0.0000025, - "input_cost_per_character_above_128k_tokens": 0.000000625, - "output_cost_per_token": 0.000005, - "output_cost_per_character": 0.00000125, - "output_cost_per_token_above_128k_tokens": 0.00001, - "output_cost_per_character_above_128k_tokens": 0.0000025, + "input_cost_per_token": 1.25e-06, + "input_cost_per_character": 3.125e-07, + "input_cost_per_image_above_128k_tokens": 0.0006575, + "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, + "input_cost_per_audio_per_second_above_128k_tokens": 6.25e-05, + "input_cost_per_token_above_128k_tokens": 2.5e-06, + "input_cost_per_character_above_128k_tokens": 6.25e-07, + "output_cost_per_token": 5e-06, + "output_cost_per_character": 1.25e-06, + "output_cost_per_token_above_128k_tokens": 1e-05, + "output_cost_per_character_above_128k_tokens": 2.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_vision": true, "supports_pdf_input": true, "supports_system_messages": true, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_response_schema": true, - "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" + "supports_tool_choice": true, + "supports_response_schema": true, + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", + "supports_parallel_function_calling": true }, "gemini-1.5-pro-002": { "max_tokens": 8192, "max_input_tokens": 2097152, "max_output_tokens": 8192, "input_cost_per_image": 0.00032875, - "input_cost_per_audio_per_second": 0.00003125, + "input_cost_per_audio_per_second": 3.125e-05, "input_cost_per_video_per_second": 0.00032875, - "input_cost_per_token": 0.00000125, - "input_cost_per_character": 0.0000003125, - "input_cost_per_image_above_128k_tokens": 0.0006575, - "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, - "input_cost_per_audio_per_second_above_128k_tokens": 0.0000625, - "input_cost_per_token_above_128k_tokens": 0.0000025, - "input_cost_per_character_above_128k_tokens": 0.000000625, - "output_cost_per_token": 0.000005, - "output_cost_per_character": 0.00000125, - "output_cost_per_token_above_128k_tokens": 0.00001, - "output_cost_per_character_above_128k_tokens": 0.0000025, + "input_cost_per_token": 1.25e-06, + "input_cost_per_character": 3.125e-07, + "input_cost_per_image_above_128k_tokens": 0.0006575, + "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, + "input_cost_per_audio_per_second_above_128k_tokens": 6.25e-05, + "input_cost_per_token_above_128k_tokens": 2.5e-06, + "input_cost_per_character_above_128k_tokens": 6.25e-07, + "output_cost_per_token": 5e-06, + "output_cost_per_character": 1.25e-06, + "output_cost_per_token_above_128k_tokens": 1e-05, + "output_cost_per_character_above_128k_tokens": 2.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_vision": true, "supports_system_messages": true, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_response_schema": true, + "supports_tool_choice": true, + "supports_response_schema": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-1.5-pro", - "deprecation_date": "2025-09-24" + "deprecation_date": "2025-09-24", + "supports_parallel_function_calling": true }, - "gemini-1.5-pro-001": { + "gemini-1.5-pro-001": { "max_tokens": 8192, "max_input_tokens": 1000000, "max_output_tokens": 8192, "input_cost_per_image": 0.00032875, - "input_cost_per_audio_per_second": 0.00003125, + "input_cost_per_audio_per_second": 3.125e-05, "input_cost_per_video_per_second": 0.00032875, - "input_cost_per_token": 0.00000125, - "input_cost_per_character": 0.0000003125, - "input_cost_per_image_above_128k_tokens": 0.0006575, - "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, - "input_cost_per_audio_per_second_above_128k_tokens": 0.0000625, - "input_cost_per_token_above_128k_tokens": 0.0000025, - "input_cost_per_character_above_128k_tokens": 0.000000625, - "output_cost_per_token": 0.000005, - "output_cost_per_character": 0.00000125, - "output_cost_per_token_above_128k_tokens": 0.00001, - "output_cost_per_character_above_128k_tokens": 0.0000025, + "input_cost_per_token": 1.25e-06, + "input_cost_per_character": 3.125e-07, + "input_cost_per_image_above_128k_tokens": 0.0006575, + "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, + "input_cost_per_audio_per_second_above_128k_tokens": 6.25e-05, + "input_cost_per_token_above_128k_tokens": 2.5e-06, + "input_cost_per_character_above_128k_tokens": 6.25e-07, + "output_cost_per_token": 5e-06, + "output_cost_per_character": 1.25e-06, + "output_cost_per_token_above_128k_tokens": 1e-05, + "output_cost_per_character_above_128k_tokens": 2.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_vision": true, "supports_system_messages": true, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_response_schema": true, + "supports_tool_choice": true, + "supports_response_schema": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "deprecation_date": "2025-05-24" + "deprecation_date": "2025-05-24", + "supports_parallel_function_calling": true }, - "gemini-1.5-pro-preview-0514": { + "gemini-1.5-pro-preview-0514": { "max_tokens": 8192, "max_input_tokens": 1000000, "max_output_tokens": 8192, "input_cost_per_image": 0.00032875, - "input_cost_per_audio_per_second": 0.00003125, + "input_cost_per_audio_per_second": 3.125e-05, "input_cost_per_video_per_second": 0.00032875, - "input_cost_per_token": 0.000000078125, - "input_cost_per_character": 0.0000003125, - "input_cost_per_image_above_128k_tokens": 0.0006575, - "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, - "input_cost_per_audio_per_second_above_128k_tokens": 0.0000625, - "input_cost_per_token_above_128k_tokens": 0.00000015625, - "input_cost_per_character_above_128k_tokens": 0.000000625, - "output_cost_per_token": 0.0000003125, - "output_cost_per_character": 0.00000125, - "output_cost_per_token_above_128k_tokens": 0.000000625, - "output_cost_per_character_above_128k_tokens": 0.0000025, + "input_cost_per_token": 7.8125e-08, + "input_cost_per_character": 3.125e-07, + "input_cost_per_image_above_128k_tokens": 0.0006575, + "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, + "input_cost_per_audio_per_second_above_128k_tokens": 6.25e-05, + "input_cost_per_token_above_128k_tokens": 1.5625e-07, + "input_cost_per_character_above_128k_tokens": 6.25e-07, + "output_cost_per_token": 3.125e-07, + "output_cost_per_character": 1.25e-06, + "output_cost_per_token_above_128k_tokens": 6.25e-07, + "output_cost_per_character_above_128k_tokens": 2.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_response_schema": true, - "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" + "supports_tool_choice": true, + "supports_response_schema": true, + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", + "supports_parallel_function_calling": true }, - "gemini-1.5-pro-preview-0215": { + "gemini-1.5-pro-preview-0215": { "max_tokens": 8192, "max_input_tokens": 1000000, "max_output_tokens": 8192, "input_cost_per_image": 0.00032875, - "input_cost_per_audio_per_second": 0.00003125, + "input_cost_per_audio_per_second": 3.125e-05, "input_cost_per_video_per_second": 0.00032875, - "input_cost_per_token": 0.000000078125, - "input_cost_per_character": 0.0000003125, - "input_cost_per_image_above_128k_tokens": 0.0006575, - "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, - "input_cost_per_audio_per_second_above_128k_tokens": 0.0000625, - "input_cost_per_token_above_128k_tokens": 0.00000015625, - "input_cost_per_character_above_128k_tokens": 0.000000625, - "output_cost_per_token": 0.0000003125, - "output_cost_per_character": 0.00000125, - "output_cost_per_token_above_128k_tokens": 0.000000625, - "output_cost_per_character_above_128k_tokens": 0.0000025, + "input_cost_per_token": 7.8125e-08, + "input_cost_per_character": 3.125e-07, + "input_cost_per_image_above_128k_tokens": 0.0006575, + "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, + "input_cost_per_audio_per_second_above_128k_tokens": 6.25e-05, + "input_cost_per_token_above_128k_tokens": 1.5625e-07, + "input_cost_per_character_above_128k_tokens": 6.25e-07, + "output_cost_per_token": 3.125e-07, + "output_cost_per_character": 1.25e-06, + "output_cost_per_token_above_128k_tokens": 6.25e-07, + "output_cost_per_character_above_128k_tokens": 2.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_response_schema": true, - "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" + "supports_tool_choice": true, + "supports_response_schema": true, + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", + "supports_parallel_function_calling": true }, "gemini-1.5-pro-preview-0409": { "max_tokens": 8192, "max_input_tokens": 1000000, "max_output_tokens": 8192, "input_cost_per_image": 0.00032875, - "input_cost_per_audio_per_second": 0.00003125, + "input_cost_per_audio_per_second": 3.125e-05, "input_cost_per_video_per_second": 0.00032875, - "input_cost_per_token": 0.000000078125, - "input_cost_per_character": 0.0000003125, - "input_cost_per_image_above_128k_tokens": 0.0006575, - "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, - "input_cost_per_audio_per_second_above_128k_tokens": 0.0000625, - "input_cost_per_token_above_128k_tokens": 0.00000015625, - "input_cost_per_character_above_128k_tokens": 0.000000625, - "output_cost_per_token": 0.0000003125, - "output_cost_per_character": 0.00000125, - "output_cost_per_token_above_128k_tokens": 0.000000625, - "output_cost_per_character_above_128k_tokens": 0.0000025, + "input_cost_per_token": 7.8125e-08, + "input_cost_per_character": 3.125e-07, + "input_cost_per_image_above_128k_tokens": 0.0006575, + "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, + "input_cost_per_audio_per_second_above_128k_tokens": 6.25e-05, + "input_cost_per_token_above_128k_tokens": 1.5625e-07, + "input_cost_per_character_above_128k_tokens": 6.25e-07, + "output_cost_per_token": 3.125e-07, + "output_cost_per_character": 1.25e-06, + "output_cost_per_token_above_128k_tokens": 6.25e-07, + "output_cost_per_character_above_128k_tokens": 2.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": true, "supports_tool_choice": true, - "supports_response_schema": true, - "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" + "supports_response_schema": true, + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", + "supports_parallel_function_calling": true }, "gemini-1.5-flash": { "max_tokens": 8192, @@ -5509,20 +5806,20 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_image": 0.00002, - "input_cost_per_video_per_second": 0.00002, - "input_cost_per_audio_per_second": 0.000002, - "input_cost_per_token": 0.000000075, - "input_cost_per_character": 0.00000001875, - "input_cost_per_token_above_128k_tokens": 0.000001, - "input_cost_per_character_above_128k_tokens": 0.00000025, - "input_cost_per_image_above_128k_tokens": 0.00004, - "input_cost_per_video_per_second_above_128k_tokens": 0.00004, - "input_cost_per_audio_per_second_above_128k_tokens": 0.000004, - "output_cost_per_token": 0.0000003, - "output_cost_per_character": 0.000000075, - "output_cost_per_token_above_128k_tokens": 0.0000006, - "output_cost_per_character_above_128k_tokens": 0.00000015, + "input_cost_per_image": 2e-05, + "input_cost_per_video_per_second": 2e-05, + "input_cost_per_audio_per_second": 2e-06, + "input_cost_per_token": 7.5e-08, + "input_cost_per_character": 1.875e-08, + "input_cost_per_token_above_128k_tokens": 1e-06, + "input_cost_per_character_above_128k_tokens": 2.5e-07, + "input_cost_per_image_above_128k_tokens": 4e-05, + "input_cost_per_video_per_second_above_128k_tokens": 4e-05, + "input_cost_per_audio_per_second_above_128k_tokens": 4e-06, + "output_cost_per_token": 3e-07, + "output_cost_per_character": 7.5e-08, + "output_cost_per_token_above_128k_tokens": 6e-07, + "output_cost_per_character_above_128k_tokens": 1.5e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -5530,7 +5827,8 @@ "supports_vision": true, "supports_response_schema": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-1.5-flash-exp-0827": { "max_tokens": 8192, @@ -5542,20 +5840,20 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_image": 0.00002, - "input_cost_per_video_per_second": 0.00002, - "input_cost_per_audio_per_second": 0.000002, - "input_cost_per_token": 0.000000004688, - "input_cost_per_character": 0.00000001875, - "input_cost_per_token_above_128k_tokens": 0.000001, - "input_cost_per_character_above_128k_tokens": 0.00000025, - "input_cost_per_image_above_128k_tokens": 0.00004, - "input_cost_per_video_per_second_above_128k_tokens": 0.00004, - "input_cost_per_audio_per_second_above_128k_tokens": 0.000004, - "output_cost_per_token": 0.0000000046875, - "output_cost_per_character": 0.00000001875, - "output_cost_per_token_above_128k_tokens": 0.000000009375, - "output_cost_per_character_above_128k_tokens": 0.0000000375, + "input_cost_per_image": 2e-05, + "input_cost_per_video_per_second": 2e-05, + "input_cost_per_audio_per_second": 2e-06, + "input_cost_per_token": 4.688e-09, + "input_cost_per_character": 1.875e-08, + "input_cost_per_token_above_128k_tokens": 1e-06, + "input_cost_per_character_above_128k_tokens": 2.5e-07, + "input_cost_per_image_above_128k_tokens": 4e-05, + "input_cost_per_video_per_second_above_128k_tokens": 4e-05, + "input_cost_per_audio_per_second_above_128k_tokens": 4e-06, + "output_cost_per_token": 4.6875e-09, + "output_cost_per_character": 1.875e-08, + "output_cost_per_token_above_128k_tokens": 9.375e-09, + "output_cost_per_character_above_128k_tokens": 3.75e-08, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -5563,7 +5861,8 @@ "supports_vision": true, "supports_response_schema": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-1.5-flash-002": { "max_tokens": 8192, @@ -5575,20 +5874,20 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_image": 0.00002, - "input_cost_per_video_per_second": 0.00002, - "input_cost_per_audio_per_second": 0.000002, - "input_cost_per_token": 0.000000075, - "input_cost_per_character": 0.00000001875, - "input_cost_per_token_above_128k_tokens": 0.000001, - "input_cost_per_character_above_128k_tokens": 0.00000025, - "input_cost_per_image_above_128k_tokens": 0.00004, - "input_cost_per_video_per_second_above_128k_tokens": 0.00004, - "input_cost_per_audio_per_second_above_128k_tokens": 0.000004, - "output_cost_per_token": 0.0000003, - "output_cost_per_character": 0.000000075, - "output_cost_per_token_above_128k_tokens": 0.0000006, - "output_cost_per_character_above_128k_tokens": 0.00000015, + "input_cost_per_image": 2e-05, + "input_cost_per_video_per_second": 2e-05, + "input_cost_per_audio_per_second": 2e-06, + "input_cost_per_token": 7.5e-08, + "input_cost_per_character": 1.875e-08, + "input_cost_per_token_above_128k_tokens": 1e-06, + "input_cost_per_character_above_128k_tokens": 2.5e-07, + "input_cost_per_image_above_128k_tokens": 4e-05, + "input_cost_per_video_per_second_above_128k_tokens": 4e-05, + "input_cost_per_audio_per_second_above_128k_tokens": 4e-06, + "output_cost_per_token": 3e-07, + "output_cost_per_character": 7.5e-08, + "output_cost_per_token_above_128k_tokens": 6e-07, + "output_cost_per_character_above_128k_tokens": 1.5e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -5597,7 +5896,8 @@ "supports_response_schema": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-1.5-flash", "deprecation_date": "2025-09-24", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-1.5-flash-001": { "max_tokens": 8192, @@ -5609,20 +5909,20 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_image": 0.00002, - "input_cost_per_video_per_second": 0.00002, - "input_cost_per_audio_per_second": 0.000002, - "input_cost_per_token": 0.000000075, - "input_cost_per_character": 0.00000001875, - "input_cost_per_token_above_128k_tokens": 0.000001, - "input_cost_per_character_above_128k_tokens": 0.00000025, - "input_cost_per_image_above_128k_tokens": 0.00004, - "input_cost_per_video_per_second_above_128k_tokens": 0.00004, - "input_cost_per_audio_per_second_above_128k_tokens": 0.000004, - "output_cost_per_token": 0.0000003, - "output_cost_per_character": 0.000000075, - "output_cost_per_token_above_128k_tokens": 0.0000006, - "output_cost_per_character_above_128k_tokens": 0.00000015, + "input_cost_per_image": 2e-05, + "input_cost_per_video_per_second": 2e-05, + "input_cost_per_audio_per_second": 2e-06, + "input_cost_per_token": 7.5e-08, + "input_cost_per_character": 1.875e-08, + "input_cost_per_token_above_128k_tokens": 1e-06, + "input_cost_per_character_above_128k_tokens": 2.5e-07, + "input_cost_per_image_above_128k_tokens": 4e-05, + "input_cost_per_video_per_second_above_128k_tokens": 4e-05, + "input_cost_per_audio_per_second_above_128k_tokens": 4e-06, + "output_cost_per_token": 3e-07, + "output_cost_per_character": 7.5e-08, + "output_cost_per_token_above_128k_tokens": 6e-07, + "output_cost_per_character_above_128k_tokens": 1.5e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -5631,7 +5931,8 @@ "supports_response_schema": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", "deprecation_date": "2025-05-24", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-1.5-flash-preview-0514": { "max_tokens": 8192, @@ -5643,27 +5944,28 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_image": 0.00002, - "input_cost_per_video_per_second": 0.00002, - "input_cost_per_audio_per_second": 0.000002, - "input_cost_per_token": 0.000000075, - "input_cost_per_character": 0.00000001875, - "input_cost_per_token_above_128k_tokens": 0.000001, - "input_cost_per_character_above_128k_tokens": 0.00000025, - "input_cost_per_image_above_128k_tokens": 0.00004, - "input_cost_per_video_per_second_above_128k_tokens": 0.00004, - "input_cost_per_audio_per_second_above_128k_tokens": 0.000004, - "output_cost_per_token": 0.0000000046875, - "output_cost_per_character": 0.00000001875, - "output_cost_per_token_above_128k_tokens": 0.000000009375, - "output_cost_per_character_above_128k_tokens": 0.0000000375, + "input_cost_per_image": 2e-05, + "input_cost_per_video_per_second": 2e-05, + "input_cost_per_audio_per_second": 2e-06, + "input_cost_per_token": 7.5e-08, + "input_cost_per_character": 1.875e-08, + "input_cost_per_token_above_128k_tokens": 1e-06, + "input_cost_per_character_above_128k_tokens": 2.5e-07, + "input_cost_per_image_above_128k_tokens": 4e-05, + "input_cost_per_video_per_second_above_128k_tokens": 4e-05, + "input_cost_per_audio_per_second_above_128k_tokens": 4e-06, + "output_cost_per_token": 4.6875e-09, + "output_cost_per_character": 1.875e-08, + "output_cost_per_token_above_128k_tokens": 9.375e-09, + "output_cost_per_character_above_128k_tokens": 3.75e-08, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, "supports_function_calling": true, "supports_vision": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-pro-experimental": { "max_tokens": 8192, @@ -5676,8 +5978,9 @@ "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": false, - "supports_tool_choice": true, - "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/gemini-experimental" + "supports_tool_choice": true, + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/gemini-experimental", + "supports_parallel_function_calling": true }, "gemini-flash-experimental": { "max_tokens": 8192, @@ -5690,8 +5993,9 @@ "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": false, - "supports_tool_choice": true, - "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/gemini-experimental" + "supports_tool_choice": true, + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/gemini-experimental", + "supports_parallel_function_calling": true }, "gemini-pro-vision": { "max_tokens": 2048, @@ -5700,15 +6004,16 @@ "max_images_per_prompt": 16, "max_videos_per_prompt": 1, "max_video_length": 2, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "input_cost_per_image": 0.0025, "litellm_provider": "vertex_ai-vision-models", "mode": "chat", "supports_function_calling": true, "supports_vision": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-1.0-pro-vision": { "max_tokens": 2048, @@ -5717,15 +6022,16 @@ "max_images_per_prompt": 16, "max_videos_per_prompt": 1, "max_video_length": 2, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "input_cost_per_image": 0.0025, "litellm_provider": "vertex_ai-vision-models", "mode": "chat", "supports_function_calling": true, "supports_vision": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-1.0-pro-vision-001": { "max_tokens": 2048, @@ -5734,8 +6040,8 @@ "max_images_per_prompt": 16, "max_videos_per_prompt": 1, "max_video_length": 2, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "input_cost_per_image": 0.0025, "litellm_provider": "vertex_ai-vision-models", "mode": "chat", @@ -5743,14 +6049,15 @@ "supports_vision": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", "deprecation_date": "2025-04-09", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "medlm-medium": { "max_tokens": 8192, "max_input_tokens": 32768, "max_output_tokens": 8192, - "input_cost_per_character": 0.0000005, - "output_cost_per_character": 0.000001, + "input_cost_per_character": 5e-07, + "output_cost_per_character": 1e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5760,8 +6067,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_character": 0.000005, - "output_cost_per_character": 0.000015, + "input_cost_per_character": 5e-06, + "output_cost_per_character": 1.5e-05, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5777,10 +6084,10 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_200k_tokens": 0.000015, + "input_cost_per_token": 1.25e-06, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_token_above_200k_tokens": 1.5e-05, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -5791,10 +6098,21 @@ "supports_pdf_input": true, "supports_response_schema": true, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], - "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing" + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing", + "supports_parallel_function_calling": true }, "gemini-2.0-pro-exp-02-05": { "max_tokens": 8192, @@ -5806,10 +6124,10 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_200k_tokens": 0.000015, + "input_cost_per_token": 1.25e-06, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_token_above_200k_tokens": 1.5e-05, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -5820,10 +6138,21 @@ "supports_pdf_input": true, "supports_response_schema": true, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], - "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing" + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing", + "supports_parallel_function_calling": true }, "gemini-2.0-flash-exp": { "max_tokens": 8192, @@ -5838,14 +6167,14 @@ "input_cost_per_image": 0, "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, - "input_cost_per_token": 0.00000015, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_token": 1.5e-07, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, - "output_cost_per_token": 0.0000006, + "output_cost_per_token": 6e-07, "output_cost_per_character": 0, "output_cost_per_token_above_128k_tokens": 0, "output_cost_per_character_above_128k_tokens": 0, @@ -5856,10 +6185,19 @@ "supports_vision": true, "supports_response_schema": true, "supports_audio_output": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-2.0-flash-001": { "max_tokens": 8192, @@ -5871,9 +6209,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.000001, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000006, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -5882,10 +6220,19 @@ "supports_response_schema": true, "supports_audio_output": true, "supports_tool_choice": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing", - "deprecation_date": "2026-02-05" + "deprecation_date": "2026-02-05", + "supports_parallel_function_calling": true }, "gemini-2.0-flash-thinking-exp": { "max_tokens": 8192, @@ -5901,9 +6248,9 @@ "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, "input_cost_per_token": 0, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, @@ -5918,10 +6265,19 @@ "supports_vision": true, "supports_response_schema": true, "supports_audio_output": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-2.0-flash", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-2.0-flash-thinking-exp-01-21": { "max_tokens": 65536, @@ -5937,9 +6293,9 @@ "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, "input_cost_per_token": 0, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, @@ -5954,10 +6310,19 @@ "supports_vision": true, "supports_response_schema": false, "supports_audio_output": false, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-2.0-flash", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini/gemini-2.5-pro-exp-03-25": { "max_tokens": 65535, @@ -5985,9 +6350,19 @@ "supports_pdf_input": true, "supports_response_schema": true, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing" }, "gemini/gemini-2.5-flash-preview-tts": { @@ -6000,10 +6375,10 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 1e-6, - "input_cost_per_token": 0.15e-6, - "output_cost_per_token": 0.6e-6, - "output_cost_per_reasoning_token": 3.5e-6, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "output_cost_per_reasoning_token": 3.5e-06, "litellm_provider": "gemini", "mode": "chat", "rpm": 10, @@ -6015,9 +6390,16 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions"], - "supported_modalities": ["text"], - "supported_output_modalities": ["audio"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions" + ], + "supported_modalities": [ + "text" + ], + "supported_output_modalities": [ + "audio" + ], "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview" }, "gemini/gemini-2.5-flash-preview-05-20": { @@ -6030,10 +6412,10 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 1e-6, - "input_cost_per_token": 0.15e-6, - "output_cost_per_token": 0.6e-6, - "output_cost_per_reasoning_token": 3.5e-6, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "output_cost_per_reasoning_token": 3.5e-06, "litellm_provider": "gemini", "mode": "chat", "rpm": 10, @@ -6045,9 +6427,19 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview" }, "gemini/gemini-2.5-flash-preview-04-17": { @@ -6060,10 +6452,10 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 1e-6, - "input_cost_per_token": 0.15e-6, - "output_cost_per_token": 0.6e-6, - "output_cost_per_reasoning_token": 3.5e-6, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "output_cost_per_reasoning_token": 3.5e-06, "litellm_provider": "gemini", "mode": "chat", "rpm": 10, @@ -6075,9 +6467,19 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview" }, "gemini-2.5-flash-preview-05-20": { @@ -6090,10 +6492,10 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 1e-6, - "input_cost_per_token": 0.15e-6, - "output_cost_per_token": 0.6e-6, - "output_cost_per_reasoning_token": 3.5e-6, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "output_cost_per_reasoning_token": 3.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_reasoning": true, @@ -6103,10 +6505,22 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions", "/v1/batch"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], - "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview" + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview", + "supports_parallel_function_calling": true }, "gemini-2.5-flash-preview-04-17": { "max_tokens": 65535, @@ -6118,10 +6532,10 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 1e-6, - "input_cost_per_token": 0.15e-6, - "output_cost_per_token": 0.6e-6, - "output_cost_per_reasoning_token": 3.5e-6, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "output_cost_per_reasoning_token": 3.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_reasoning": true, @@ -6131,10 +6545,22 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions", "/v1/batch"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], - "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview" + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview", + "supports_parallel_function_calling": true }, "gemini-2.0-flash": { "max_tokens": 8192, @@ -6146,9 +6572,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000004, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -6157,10 +6583,19 @@ "supports_response_schema": true, "supports_audio_output": true, "supports_audio_input": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "supports_tool_choice": true, - "source": "https://ai.google.dev/pricing#2_0flash" + "source": "https://ai.google.dev/pricing#2_0flash", + "supports_parallel_function_calling": true }, "gemini-2.0-flash-lite": { "max_input_tokens": 1048576, @@ -6171,9 +6606,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 50, - "input_cost_per_audio_token": 0.000000075, - "input_cost_per_token": 0.000000075, - "output_cost_per_token": 0.0000003, + "input_cost_per_audio_token": 7.5e-08, + "input_cost_per_token": 7.5e-08, + "output_cost_per_token": 3e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -6181,10 +6616,18 @@ "supports_vision": true, "supports_response_schema": true, "supports_audio_output": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-2.0-flash", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-2.0-flash-lite-001": { "max_input_tokens": 1048576, @@ -6195,9 +6638,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 50, - "input_cost_per_audio_token": 0.000000075, - "input_cost_per_token": 0.000000075, - "output_cost_per_token": 0.0000003, + "input_cost_per_audio_token": 7.5e-08, + "input_cost_per_token": 7.5e-08, + "output_cost_per_token": 3e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -6205,11 +6648,19 @@ "supports_vision": true, "supports_response_schema": true, "supports_audio_output": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-2.0-flash", "supports_tool_choice": true, - "deprecation_date": "2026-02-25" + "deprecation_date": "2026-02-25", + "supports_parallel_function_calling": true }, "gemini-2.5-pro-preview-05-06": { "max_tokens": 65535, @@ -6221,11 +6672,11 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.00000125, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_200k_tokens": 0.000015, + "input_cost_per_audio_token": 1.25e-06, + "input_cost_per_token": 1.25e-06, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_token_above_200k_tokens": 1.5e-05, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_reasoning": true, @@ -6235,10 +6686,22 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions", "/v1/batch"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], - "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview" + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview", + "supports_parallel_function_calling": true }, "gemini-2.5-pro-preview-03-25": { "max_tokens": 65535, @@ -6250,11 +6713,11 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.00000125, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_200k_tokens": 0.000015, + "input_cost_per_audio_token": 1.25e-06, + "input_cost_per_token": 1.25e-06, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_token_above_200k_tokens": 1.5e-05, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_reasoning": true, @@ -6264,10 +6727,22 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions", "/v1/batch"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], - "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview" + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview", + "supports_parallel_function_calling": true }, "gemini-2.0-flash-preview-image-generation": { "max_tokens": 8192, @@ -6279,9 +6754,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000004, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -6290,10 +6765,19 @@ "supports_response_schema": true, "supports_audio_output": true, "supports_audio_input": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "supports_tool_choice": true, - "source": "https://ai.google.dev/pricing#2_0flash" + "source": "https://ai.google.dev/pricing#2_0flash", + "supports_parallel_function_calling": true }, "gemini-2.5-pro-preview-tts": { "max_tokens": 65535, @@ -6305,11 +6789,11 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_200k_tokens": 0.000015, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1.25e-06, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_token_above_200k_tokens": 1.5e-05, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -6318,9 +6802,14 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_modalities": ["text"], - "supported_output_modalities": ["audio"], - "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.5-pro-preview" + "supported_modalities": [ + "text" + ], + "supported_output_modalities": [ + "audio" + ], + "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.5-pro-preview", + "supports_parallel_function_calling": true }, "gemini/gemini-2.0-pro-exp-02-05": { "max_tokens": 8192, @@ -6336,9 +6825,9 @@ "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, "input_cost_per_token": 0, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, @@ -6370,9 +6859,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000004, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "gemini", "mode": "chat", "rpm": 10000, @@ -6383,8 +6872,16 @@ "supports_response_schema": true, "supports_audio_output": true, "supports_audio_input": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "supports_tool_choice": true, "source": "https://ai.google.dev/pricing#2_0flash" }, @@ -6398,9 +6895,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000004, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "gemini", "mode": "chat", "rpm": 10000, @@ -6411,8 +6908,16 @@ "supports_response_schema": true, "supports_audio_output": true, "supports_audio_input": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "supports_tool_choice": true, "source": "https://ai.google.dev/pricing#2_0flash" }, @@ -6425,9 +6930,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 50, - "input_cost_per_audio_token": 0.000000075, - "input_cost_per_token": 0.000000075, - "output_cost_per_token": 0.0000003, + "input_cost_per_audio_token": 7.5e-08, + "input_cost_per_token": 7.5e-08, + "output_cost_per_token": 3e-07, "litellm_provider": "gemini", "mode": "chat", "tpm": 4000000, @@ -6438,8 +6943,15 @@ "supports_response_schema": true, "supports_audio_output": true, "supports_tool_choice": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.0-flash-lite" }, "gemini/gemini-2.0-flash-001": { @@ -6452,9 +6964,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000004, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "gemini", "mode": "chat", "rpm": 10000, @@ -6465,8 +6977,16 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "source": "https://ai.google.dev/pricing#2_0flash" }, "gemini/gemini-2.5-pro-preview-tts": { @@ -6479,11 +6999,11 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_200k_tokens": 0.000015, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1.25e-06, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_token_above_200k_tokens": 1.5e-05, "litellm_provider": "gemini", "mode": "chat", "rpm": 10000, @@ -6494,8 +7014,12 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_modalities": ["text"], - "supported_output_modalities": ["audio"], + "supported_modalities": [ + "text" + ], + "supported_output_modalities": [ + "audio" + ], "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.5-pro-preview" }, "gemini/gemini-2.5-pro-preview-05-06": { @@ -6508,11 +7032,11 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_200k_tokens": 0.000015, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1.25e-06, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_token_above_200k_tokens": 1.5e-05, "litellm_provider": "gemini", "mode": "chat", "rpm": 10000, @@ -6523,8 +7047,15 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.5-pro-preview" }, "gemini/gemini-2.5-pro-preview-03-25": { @@ -6537,11 +7068,11 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_200k_tokens": 0.000015, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1.25e-06, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_token_above_200k_tokens": 1.5e-05, "litellm_provider": "gemini", "mode": "chat", "rpm": 10000, @@ -6552,8 +7083,15 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.5-pro-preview" }, "gemini/gemini-2.0-flash-exp": { @@ -6570,9 +7108,9 @@ "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, "input_cost_per_token": 0, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, @@ -6589,8 +7127,16 @@ "supports_audio_output": true, "tpm": 4000000, "rpm": 10, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-2.0-flash", "supports_tool_choice": true }, @@ -6604,9 +7150,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.000000075, - "input_cost_per_token": 0.000000075, - "output_cost_per_token": 0.0000003, + "input_cost_per_audio_token": 7.5e-08, + "input_cost_per_token": 7.5e-08, + "output_cost_per_token": 3e-07, "litellm_provider": "gemini", "mode": "chat", "rpm": 60000, @@ -6617,8 +7163,15 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-2.0-flash-lite" }, "gemini/gemini-2.0-flash-thinking-exp": { @@ -6635,9 +7188,9 @@ "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, "input_cost_per_token": 0, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, @@ -6654,8 +7207,16 @@ "supports_audio_output": true, "tpm": 4000000, "rpm": 10, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-2.0-flash", "supports_tool_choice": true }, @@ -6673,9 +7234,9 @@ "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, "input_cost_per_token": 0, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, @@ -6692,8 +7253,16 @@ "supports_audio_output": true, "tpm": 4000000, "rpm": 10, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-2.0-flash", "supports_tool_choice": true }, @@ -6705,9 +7274,9 @@ "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, "input_cost_per_token": 0, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, @@ -6733,9 +7302,9 @@ "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, "input_cost_per_token": 0, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, @@ -6757,8 +7326,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6770,8 +7339,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6784,8 +7353,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6798,8 +7367,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6813,8 +7382,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6828,8 +7397,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6843,10 +7412,10 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6864,15 +7433,15 @@ "max_tokens": 32000, "max_input_tokens": 200000, "max_output_tokens": 32000, - "input_cost_per_token": 15e-6, - "output_cost_per_token": 75e-6, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 18.75e-6, - "cache_read_input_token_cost": 1.5e-6, + "cache_creation_input_token_cost": 1.875e-05, + "cache_read_input_token_cost": 1.5e-06, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6890,15 +7459,15 @@ "max_tokens": 64000, "max_input_tokens": 200000, "max_output_tokens": 64000, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 15e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 3.75e-6, - "cache_read_input_token_cost": 0.3e-6, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6913,11 +7482,11 @@ "supports_computer_use": true }, "vertex_ai/claude-3-haiku": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6926,11 +7495,11 @@ "supports_tool_choice": true }, "vertex_ai/claude-3-haiku@20240307": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6942,8 +7511,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000005, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 5e-06, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6955,8 +7524,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000005, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 5e-06, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6968,8 +7537,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6981,8 +7550,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -7002,60 +7571,84 @@ "supports_tool_choice": true }, "vertex_ai/meta/llama-4-scout-17b-16e-instruct-maas": { - "max_tokens": 10e6, - "max_input_tokens": 10e6, - "max_output_tokens": 10e6, - "input_cost_per_token": 0.25e-6, - "output_cost_per_token": 0.70e-6, + "max_tokens": 10000000.0, + "max_input_tokens": 10000000.0, + "max_output_tokens": 10000000.0, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "vertex_ai-llama_models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#partner-models", "supports_tool_choice": true, "supports_function_calling": true, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text", "code"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] }, "vertex_ai/meta/llama-4-scout-17b-128e-instruct-maas": { - "max_tokens": 10e6, - "max_input_tokens": 10e6, - "max_output_tokens": 10e6, - "input_cost_per_token": 0.25e-6, - "output_cost_per_token": 0.70e-6, + "max_tokens": 10000000.0, + "max_input_tokens": 10000000.0, + "max_output_tokens": 10000000.0, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "vertex_ai-llama_models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#partner-models", "supports_tool_choice": true, "supports_function_calling": true, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text", "code"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] }, "vertex_ai/meta/llama-4-maverick-17b-128e-instruct-maas": { - "max_tokens": 1e6, - "max_input_tokens": 1e6, - "max_output_tokens": 1e6, - "input_cost_per_token": 0.35e-6, - "output_cost_per_token": 1.15e-6, + "max_tokens": 1000000.0, + "max_input_tokens": 1000000.0, + "max_output_tokens": 1000000.0, + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 1.15e-06, "litellm_provider": "vertex_ai-llama_models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#partner-models", "supports_tool_choice": true, "supports_function_calling": true, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text", "code"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] }, "vertex_ai/meta/llama-4-maverick-17b-16e-instruct-maas": { - "max_tokens": 1e6, - "max_input_tokens": 1e6, - "max_output_tokens": 1e6, - "input_cost_per_token": 0.35e-6, - "output_cost_per_token": 1.15e-6, + "max_tokens": 1000000.0, + "max_input_tokens": 1000000.0, + "max_output_tokens": 1000000.0, + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 1.15e-06, "litellm_provider": "vertex_ai-llama_models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#partner-models", "supports_tool_choice": true, "supports_function_calling": true, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text", "code"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] }, "vertex_ai/meta/llama3-70b-instruct-maas": { "max_tokens": 32000, @@ -7096,8 +7689,8 @@ "max_tokens": 8191, "max_input_tokens": 128000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7107,8 +7700,8 @@ "max_tokens": 8191, "max_input_tokens": 128000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7118,8 +7711,8 @@ "max_tokens": 8191, "max_input_tokens": 128000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7129,8 +7722,8 @@ "max_tokens": 8191, "max_input_tokens": 128000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7140,8 +7733,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7151,8 +7744,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "vertex_ai-mistral_models", "supports_function_calling": true, "mode": "chat", @@ -7162,8 +7755,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7174,8 +7767,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "vertex_ai-ai21_models", "mode": "chat", "supports_tool_choice": true @@ -7184,8 +7777,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000008, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, "litellm_provider": "vertex_ai-ai21_models", "mode": "chat", "supports_tool_choice": true @@ -7194,8 +7787,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "vertex_ai-ai21_models", "mode": "chat", "supports_tool_choice": true @@ -7204,8 +7797,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "vertex_ai-ai21_models", "mode": "chat", "supports_tool_choice": true @@ -7214,8 +7807,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000008, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, "litellm_provider": "vertex_ai-ai21_models", "mode": "chat", "supports_tool_choice": true @@ -7224,8 +7817,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7235,8 +7828,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7246,8 +7839,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7257,15 +7850,15 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, "supports_tool_choice": true }, "vertex_ai/imagegeneration@006": { - "output_cost_per_image": 0.020, + "output_cost_per_image": 0.02, "litellm_provider": "vertex_ai-image-models", "mode": "image_generation", "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing" @@ -7292,8 +7885,8 @@ "max_tokens": 2048, "max_input_tokens": 2048, "output_vector_size": 768, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7303,8 +7896,8 @@ "max_tokens": 2048, "max_input_tokens": 2048, "output_vector_size": 768, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7314,8 +7907,8 @@ "max_tokens": 2048, "max_input_tokens": 2048, "output_vector_size": 768, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7325,42 +7918,54 @@ "max_tokens": 2048, "max_input_tokens": 2048, "output_vector_size": 768, - "input_cost_per_character": 0.0000002, + "input_cost_per_character": 2e-07, "input_cost_per_image": 0.0001, "input_cost_per_video_per_second": 0.0005, - "input_cost_per_video_per_second_above_8s_interval": 0.0010, - "input_cost_per_video_per_second_above_15s_interval": 0.0020, - "input_cost_per_token": 0.0000008, + "input_cost_per_video_per_second_above_8s_interval": 0.001, + "input_cost_per_video_per_second_above_15s_interval": 0.002, + "input_cost_per_token": 8e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", - "supported_endpoints": ["/v1/embeddings"], - "supported_modalities": ["text", "image", "video"], + "supported_endpoints": [ + "/v1/embeddings" + ], + "supported_modalities": [ + "text", + "image", + "video" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models" }, "multimodalembedding@001": { "max_tokens": 2048, "max_input_tokens": 2048, "output_vector_size": 768, - "input_cost_per_character": 0.0000002, + "input_cost_per_character": 2e-07, "input_cost_per_image": 0.0001, "input_cost_per_video_per_second": 0.0005, - "input_cost_per_video_per_second_above_8s_interval": 0.0010, - "input_cost_per_video_per_second_above_15s_interval": 0.0020, - "input_cost_per_token": 0.0000008, + "input_cost_per_video_per_second_above_8s_interval": 0.001, + "input_cost_per_video_per_second_above_15s_interval": 0.002, + "input_cost_per_token": 8e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", - "supported_endpoints": ["/v1/embeddings"], - "supported_modalities": ["text", "image", "video"], + "supported_endpoints": [ + "/v1/embeddings" + ], + "supported_modalities": [ + "text", + "image", + "video" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models" }, "text-embedding-large-exp-03-07": { "max_tokens": 8192, "max_input_tokens": 8192, "output_vector_size": 3072, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7370,8 +7975,8 @@ "max_tokens": 3072, "max_input_tokens": 3072, "output_vector_size": 768, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7381,8 +7986,8 @@ "max_tokens": 3072, "max_input_tokens": 3072, "output_vector_size": 768, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7392,8 +7997,8 @@ "max_tokens": 3072, "max_input_tokens": 3072, "output_vector_size": 768, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7403,8 +8008,8 @@ "max_tokens": 3072, "max_input_tokens": 3072, "output_vector_size": 768, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7414,8 +8019,8 @@ "max_tokens": 3072, "max_input_tokens": 3072, "output_vector_size": 768, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7425,18 +8030,18 @@ "max_tokens": 3072, "max_input_tokens": 3072, "output_vector_size": 768, - "input_cost_per_token": 0.00000000625, - "input_cost_per_token_batch_requests": 0.000000005, + "input_cost_per_token": 6.25e-09, + "input_cost_per_token_batch_requests": 5e-09, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing" }, - "text-multilingual-embedding-preview-0409":{ + "text-multilingual-embedding-preview-0409": { "max_tokens": 3072, "max_input_tokens": 3072, "output_vector_size": 768, - "input_cost_per_token": 0.00000000625, + "input_cost_per_token": 6.25e-09, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7446,8 +8051,8 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "palm", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -7456,8 +8061,8 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "palm", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -7466,8 +8071,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "palm", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -7476,8 +8081,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "palm", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -7486,8 +8091,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "palm", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -7496,8 +8101,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "palm", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -7511,13 +8116,13 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, - "cache_read_input_token_cost": 0.00000001875, - "cache_creation_input_token_cost": 0.000001, - "input_cost_per_token": 0.000000075, - "input_cost_per_token_above_128k_tokens": 0.00000015, - "output_cost_per_token": 0.0000003, - "output_cost_per_token_above_128k_tokens": 0.0000006, + "max_pdf_size_mb": 30, + "cache_read_input_token_cost": 1.875e-08, + "cache_creation_input_token_cost": 1e-06, + "input_cost_per_token": 7.5e-08, + "input_cost_per_token_above_128k_tokens": 1.5e-07, + "output_cost_per_token": 3e-07, + "output_cost_per_token_above_128k_tokens": 6e-07, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, @@ -7540,13 +8145,13 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, - "cache_read_input_token_cost": 0.00000001875, - "cache_creation_input_token_cost": 0.000001, - "input_cost_per_token": 0.000000075, - "input_cost_per_token_above_128k_tokens": 0.00000015, - "output_cost_per_token": 0.0000003, - "output_cost_per_token_above_128k_tokens": 0.0000006, + "max_pdf_size_mb": 30, + "cache_read_input_token_cost": 1.875e-08, + "cache_creation_input_token_cost": 1e-06, + "input_cost_per_token": 7.5e-08, + "input_cost_per_token_above_128k_tokens": 1.5e-07, + "output_cost_per_token": 3e-07, + "output_cost_per_token_above_128k_tokens": 6e-07, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, @@ -7569,17 +8174,17 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, - "input_cost_per_token": 0.000000075, - "input_cost_per_token_above_128k_tokens": 0.00000015, - "output_cost_per_token": 0.0000003, - "output_cost_per_token_above_128k_tokens": 0.0000006, + "max_pdf_size_mb": 30, + "input_cost_per_token": 7.5e-08, + "input_cost_per_token_above_128k_tokens": 1.5e-07, + "output_cost_per_token": 3e-07, + "output_cost_per_token_above_128k_tokens": 6e-07, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, "supports_function_calling": true, "supports_vision": true, - "supports_response_schema": true, + "supports_response_schema": true, "tpm": 4000000, "rpm": 2000, "source": "https://ai.google.dev/pricing", @@ -7594,11 +8199,11 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, - "input_cost_per_token": 0.000000075, - "input_cost_per_token_above_128k_tokens": 0.00000015, - "output_cost_per_token": 0.0000003, - "output_cost_per_token_above_128k_tokens": 0.0000006, + "max_pdf_size_mb": 30, + "input_cost_per_token": 7.5e-08, + "input_cost_per_token_above_128k_tokens": 1.5e-07, + "output_cost_per_token": 3e-07, + "output_cost_per_token_above_128k_tokens": 6e-07, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, @@ -7620,7 +8225,7 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, + "max_pdf_size_mb": 30, "input_cost_per_token": 0, "input_cost_per_token_above_128k_tokens": 0, "output_cost_per_token": 0, @@ -7646,7 +8251,7 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, + "max_pdf_size_mb": 30, "input_cost_per_token": 0, "input_cost_per_token_above_128k_tokens": 0, "output_cost_per_token": 0, @@ -7672,7 +8277,7 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, + "max_pdf_size_mb": 30, "input_cost_per_token": 0, "input_cost_per_token_above_128k_tokens": 0, "output_cost_per_token": 0, @@ -7701,7 +8306,7 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, + "max_pdf_size_mb": 30, "input_cost_per_token": 0, "input_cost_per_token_above_128k_tokens": 0, "output_cost_per_token": 0, @@ -7730,7 +8335,7 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, + "max_pdf_size_mb": 30, "input_cost_per_token": 0, "input_cost_per_token_above_128k_tokens": 0, "output_cost_per_token": 0, @@ -7755,7 +8360,7 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, + "max_pdf_size_mb": 30, "input_cost_per_token": 0, "input_cost_per_token_above_128k_tokens": 0, "output_cost_per_token": 0, @@ -7775,10 +8380,10 @@ "max_tokens": 8192, "max_input_tokens": 32760, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000035, - "input_cost_per_token_above_128k_tokens": 0.0000007, - "output_cost_per_token": 0.00000105, - "output_cost_per_token_above_128k_tokens": 0.0000021, + "input_cost_per_token": 3.5e-07, + "input_cost_per_token_above_128k_tokens": 7e-07, + "output_cost_per_token": 1.05e-06, + "output_cost_per_token_above_128k_tokens": 2.1e-06, "litellm_provider": "gemini", "mode": "chat", "supports_function_calling": true, @@ -7792,17 +8397,17 @@ "max_tokens": 8192, "max_input_tokens": 2097152, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000035, - "input_cost_per_token_above_128k_tokens": 0.000007, - "output_cost_per_token": 0.0000105, - "output_cost_per_token_above_128k_tokens": 0.000021, + "input_cost_per_token": 3.5e-06, + "input_cost_per_token_above_128k_tokens": 7e-06, + "output_cost_per_token": 1.05e-05, + "output_cost_per_token_above_128k_tokens": 2.1e-05, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, "supports_function_calling": true, "supports_vision": true, - "supports_tool_choice": true, - "supports_response_schema": true, + "supports_tool_choice": true, + "supports_response_schema": true, "tpm": 4000000, "rpm": 1000, "source": "https://ai.google.dev/pricing" @@ -7811,17 +8416,17 @@ "max_tokens": 8192, "max_input_tokens": 2097152, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000035, - "input_cost_per_token_above_128k_tokens": 0.000007, - "output_cost_per_token": 0.0000105, - "output_cost_per_token_above_128k_tokens": 0.000021, + "input_cost_per_token": 3.5e-06, + "input_cost_per_token_above_128k_tokens": 7e-06, + "output_cost_per_token": 1.05e-05, + "output_cost_per_token_above_128k_tokens": 2.1e-05, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, "supports_function_calling": true, "supports_vision": true, - "supports_tool_choice": true, - "supports_response_schema": true, + "supports_tool_choice": true, + "supports_response_schema": true, "supports_prompt_caching": true, "tpm": 4000000, "rpm": 1000, @@ -7832,17 +8437,17 @@ "max_tokens": 8192, "max_input_tokens": 2097152, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000035, - "input_cost_per_token_above_128k_tokens": 0.000007, - "output_cost_per_token": 0.0000105, - "output_cost_per_token_above_128k_tokens": 0.000021, + "input_cost_per_token": 3.5e-06, + "input_cost_per_token_above_128k_tokens": 7e-06, + "output_cost_per_token": 1.05e-05, + "output_cost_per_token_above_128k_tokens": 2.1e-05, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, "supports_function_calling": true, "supports_vision": true, - "supports_tool_choice": true, - "supports_response_schema": true, + "supports_tool_choice": true, + "supports_response_schema": true, "supports_prompt_caching": true, "tpm": 4000000, "rpm": 1000, @@ -7853,10 +8458,10 @@ "max_tokens": 8192, "max_input_tokens": 2097152, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000035, - "input_cost_per_token_above_128k_tokens": 0.000007, - "output_cost_per_token": 0.0000105, - "output_cost_per_token_above_128k_tokens": 0.000021, + "input_cost_per_token": 3.5e-06, + "input_cost_per_token_above_128k_tokens": 7e-06, + "output_cost_per_token": 1.05e-05, + "output_cost_per_token_above_128k_tokens": 2.1e-05, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, @@ -7891,17 +8496,17 @@ "max_tokens": 8192, "max_input_tokens": 1048576, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000035, - "input_cost_per_token_above_128k_tokens": 0.000007, - "output_cost_per_token": 0.00000105, - "output_cost_per_token_above_128k_tokens": 0.000021, + "input_cost_per_token": 3.5e-06, + "input_cost_per_token_above_128k_tokens": 7e-06, + "output_cost_per_token": 1.05e-06, + "output_cost_per_token_above_128k_tokens": 2.1e-05, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, "supports_function_calling": true, "supports_vision": true, - "supports_tool_choice": true, - "supports_response_schema": true, + "supports_tool_choice": true, + "supports_response_schema": true, "tpm": 4000000, "rpm": 1000, "source": "https://ai.google.dev/pricing" @@ -7910,10 +8515,10 @@ "max_tokens": 2048, "max_input_tokens": 30720, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000035, - "input_cost_per_token_above_128k_tokens": 0.0000007, - "output_cost_per_token": 0.00000105, - "output_cost_per_token_above_128k_tokens": 0.0000021, + "input_cost_per_token": 3.5e-07, + "input_cost_per_token_above_128k_tokens": 7e-07, + "output_cost_per_token": 1.05e-06, + "output_cost_per_token_above_128k_tokens": 2.1e-06, "litellm_provider": "gemini", "mode": "chat", "supports_function_calling": true, @@ -7927,8 +8532,8 @@ "gemini/gemini-gemma-2-27b-it": { "max_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000035, - "output_cost_per_token": 0.00000105, + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 1.05e-06, "litellm_provider": "gemini", "mode": "chat", "supports_function_calling": true, @@ -7939,8 +8544,8 @@ "gemini/gemini-gemma-2-9b-it": { "max_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000035, - "output_cost_per_token": 0.00000105, + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 1.05e-06, "litellm_provider": "gemini", "mode": "chat", "supports_function_calling": true, @@ -7952,8 +8557,8 @@ "max_tokens": 8000, "max_input_tokens": 256000, "max_output_tokens": 8000, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "cohere_chat", "mode": "chat", "supports_function_calling": true, @@ -7963,8 +8568,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "cohere_chat", "mode": "chat", "supports_function_calling": true, @@ -7974,8 +8579,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "cohere_chat", "mode": "chat", "supports_function_calling": true, @@ -7985,8 +8590,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000000375, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 3.75e-08, "litellm_provider": "cohere_chat", "mode": "chat", "supports_function_calling": true, @@ -7997,8 +8602,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "cohere_chat", "mode": "chat", "supports_tool_choice": true @@ -8007,8 +8612,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "cohere_chat", "mode": "chat", "supports_function_calling": true, @@ -8018,28 +8623,28 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "cohere_chat", "mode": "chat", "supports_function_calling": true, "supports_tool_choice": true }, "command-nightly": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "cohere", "mode": "completion" }, - "command": { - "max_tokens": 4096, + "command": { + "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "cohere", "mode": "completion" }, @@ -8099,52 +8704,52 @@ "mode": "rerank" }, "embed-english-light-v3.0": { - "max_tokens": 1024, + "max_tokens": 1024, "max_input_tokens": 1024, - "input_cost_per_token": 0.00000010, - "output_cost_per_token": 0.00000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "cohere", "mode": "embedding" }, "embed-multilingual-v3.0": { - "max_tokens": 1024, + "max_tokens": 1024, "max_input_tokens": 1024, - "input_cost_per_token": 0.00000010, - "output_cost_per_token": 0.00000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "cohere", "supports_embedding_image_input": true, "mode": "embedding" }, "embed-english-v2.0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 4096, - "input_cost_per_token": 0.00000010, - "output_cost_per_token": 0.00000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "cohere", "mode": "embedding" }, "embed-english-light-v2.0": { - "max_tokens": 1024, + "max_tokens": 1024, "max_input_tokens": 1024, - "input_cost_per_token": 0.00000010, - "output_cost_per_token": 0.00000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "cohere", "mode": "embedding" }, "embed-multilingual-v2.0": { - "max_tokens": 768, + "max_tokens": 768, "max_input_tokens": 768, - "input_cost_per_token": 0.00000010, - "output_cost_per_token": 0.00000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "cohere", "mode": "embedding" }, "embed-english-v3.0": { - "max_tokens": 1024, + "max_tokens": 1024, "max_input_tokens": 1024, - "input_cost_per_token": 0.00000010, + "input_cost_per_token": 1e-07, "input_cost_per_image": 0.0001, - "output_cost_per_token": 0.00000, + "output_cost_per_token": 0.0, "litellm_provider": "cohere", "mode": "embedding", "supports_image_input": true, @@ -8157,8 +8762,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8167,8 +8772,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8177,8 +8782,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000065, - "output_cost_per_token": 0.00000275, + "input_cost_per_token": 6.5e-07, + "output_cost_per_token": 2.75e-06, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8187,8 +8792,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000065, - "output_cost_per_token": 0.00000275, + "input_cost_per_token": 6.5e-07, + "output_cost_per_token": 2.75e-06, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8197,8 +8802,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000005, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 5e-08, + "output_cost_per_token": 2.5e-07, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8207,8 +8812,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000005, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 5e-08, + "output_cost_per_token": 2.5e-07, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8217,8 +8822,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000065, - "output_cost_per_token": 0.00000275, + "input_cost_per_token": 6.5e-07, + "output_cost_per_token": 2.75e-06, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8227,8 +8832,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000065, - "output_cost_per_token": 0.00000275, + "input_cost_per_token": 6.5e-07, + "output_cost_per_token": 2.75e-06, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8237,8 +8842,8 @@ "max_tokens": 8086, "max_input_tokens": 8086, "max_output_tokens": 8086, - "input_cost_per_token": 0.00000005, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 5e-08, + "output_cost_per_token": 2.5e-07, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8247,8 +8852,8 @@ "max_tokens": 8086, "max_input_tokens": 8086, "max_output_tokens": 8086, - "input_cost_per_token": 0.00000005, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 5e-08, + "output_cost_per_token": 2.5e-07, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8257,8 +8862,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000005, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 5e-08, + "output_cost_per_token": 2.5e-07, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8267,8 +8872,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000005, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 5e-08, + "output_cost_per_token": 2.5e-07, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8277,8 +8882,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.000001, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 1e-06, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8287,12 +8892,12 @@ "max_tokens": 8192, "max_input_tokens": 65336, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000055, - "input_cost_per_token_cache_hit": 0.00000014, - "output_cost_per_token": 0.00000219, + "input_cost_per_token": 5.5e-07, + "input_cost_per_token_cache_hit": 1.4e-07, + "output_cost_per_token": 2.19e-06, "litellm_provider": "openrouter", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_assistant_prefill": true, "supports_reasoning": true, "supports_tool_choice": true, @@ -8302,8 +8907,8 @@ "max_tokens": 8192, "max_input_tokens": 65536, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000014, - "output_cost_per_token": 0.00000028, + "input_cost_per_token": 1.4e-07, + "output_cost_per_token": 2.8e-07, "litellm_provider": "openrouter", "supports_prompt_caching": true, "mode": "chat", @@ -8313,8 +8918,8 @@ "max_tokens": 8192, "max_input_tokens": 66000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000014, - "output_cost_per_token": 0.00000028, + "input_cost_per_token": 1.4e-07, + "output_cost_per_token": 2.8e-07, "litellm_provider": "openrouter", "supports_prompt_caching": true, "mode": "chat", @@ -8322,8 +8927,8 @@ }, "openrouter/microsoft/wizardlm-2-8x22b:nitro": { "max_tokens": 65536, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000001, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true @@ -8332,9 +8937,9 @@ "max_tokens": 8192, "max_input_tokens": 1000000, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.0000075, - "input_cost_per_image": 0.00265, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 7.5e-06, + "input_cost_per_image": 0.00265, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8351,9 +8956,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000004, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_system_messages": true, @@ -8365,33 +8970,33 @@ }, "openrouter/mistralai/mixtral-8x22b-instruct": { "max_tokens": 65536, - "input_cost_per_token": 0.00000065, - "output_cost_per_token": 0.00000065, + "input_cost_per_token": 6.5e-07, + "output_cost_per_token": 6.5e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/cohere/command-r-plus": { "max_tokens": 128000, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/databricks/dbrx-instruct": { "max_tokens": 32768, - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/anthropic/claude-3-haiku": { "max_tokens": 200000, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, - "input_cost_per_image": 0.0004, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, + "input_cost_per_image": 0.0004, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8400,8 +9005,8 @@ }, "openrouter/anthropic/claude-3-5-haiku": { "max_tokens": 200000, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000005, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 5e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8411,8 +9016,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8424,8 +9029,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000005, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 5e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8437,8 +9042,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8452,8 +9057,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8466,8 +9071,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "input_cost_per_image": 0.0048, "litellm_provider": "openrouter", "mode": "chat", @@ -8483,8 +9088,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "input_cost_per_image": 0.0048, "litellm_provider": "openrouter", "mode": "chat", @@ -8496,9 +9101,9 @@ }, "openrouter/anthropic/claude-3-sonnet": { "max_tokens": 200000, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "input_cost_per_image": 0.0048, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "input_cost_per_image": 0.0048, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8507,33 +9112,33 @@ }, "openrouter/mistralai/mistral-large": { "max_tokens": 32000, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "mistralai/mistral-small-3.1-24b-instruct": { "max_tokens": 32000, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000003, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 3e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/cognitivecomputations/dolphin-mixtral-8x7b": { "max_tokens": 32769, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/google/gemini-pro-vision": { "max_tokens": 45875, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000375, - "input_cost_per_image": 0.0025, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 3.75e-07, + "input_cost_per_image": 0.0025, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8542,8 +9147,8 @@ }, "openrouter/fireworks/firellava-13b": { "max_tokens": 4096, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true @@ -8558,24 +9163,24 @@ }, "openrouter/meta-llama/llama-3-8b-instruct:extended": { "max_tokens": 16384, - "input_cost_per_token": 0.000000225, - "output_cost_per_token": 0.00000225, + "input_cost_per_token": 2.25e-07, + "output_cost_per_token": 2.25e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/meta-llama/llama-3-70b-instruct:nitro": { "max_tokens": 8192, - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/meta-llama/llama-3-70b-instruct": { "max_tokens": 8192, - "input_cost_per_token": 0.00000059, - "output_cost_per_token": 0.00000079, + "input_cost_per_token": 5.9e-07, + "output_cost_per_token": 7.9e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true @@ -8584,9 +9189,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.00006, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8601,8 +9206,8 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000012, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.2e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8614,8 +9219,8 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000012, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.2e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8627,8 +9232,8 @@ "max_tokens": 32768, "max_input_tokens": 128000, "max_output_tokens": 32768, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8640,8 +9245,8 @@ "max_tokens": 32768, "max_input_tokens": 128000, "max_output_tokens": 32768, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8653,8 +9258,8 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.0000044, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8667,8 +9272,8 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.0000044, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8681,8 +9286,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000010, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8694,8 +9299,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8705,9 +9310,9 @@ }, "openrouter/openai/gpt-4-vision-preview": { "max_tokens": 130000, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, - "input_cost_per_image": 0.01445, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, + "input_cost_per_image": 0.01445, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8716,24 +9321,24 @@ }, "openrouter/openai/gpt-3.5-turbo": { "max_tokens": 4095, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/openai/gpt-3.5-turbo-16k": { "max_tokens": 16383, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000004, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 4e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/openai/gpt-4": { "max_tokens": 8192, - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.00006, + "input_cost_per_token": 3e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true @@ -8741,8 +9346,8 @@ "openrouter/anthropic/claude-instant-v1": { "max_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000163, - "output_cost_per_token": 0.00000551, + "input_cost_per_token": 1.63e-06, + "output_cost_per_token": 5.51e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true @@ -8750,8 +9355,8 @@ "openrouter/anthropic/claude-2": { "max_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00001102, - "output_cost_per_token": 0.00003268, + "input_cost_per_token": 1.102e-05, + "output_cost_per_token": 3.268e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true @@ -8760,8 +9365,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8771,96 +9376,96 @@ }, "openrouter/google/palm-2-chat-bison": { "max_tokens": 25804, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/google/palm-2-codechat-bison": { "max_tokens": 20070, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/meta-llama/llama-2-13b-chat": { "max_tokens": 4096, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/meta-llama/llama-2-70b-chat": { "max_tokens": 4096, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 1.5e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/meta-llama/codellama-34b-instruct": { "max_tokens": 8192, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/nousresearch/nous-hermes-llama2-13b": { "max_tokens": 4096, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/mancer/weaver": { "max_tokens": 8000, - "input_cost_per_token": 0.000005625, - "output_cost_per_token": 0.000005625, + "input_cost_per_token": 5.625e-06, + "output_cost_per_token": 5.625e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/gryphe/mythomax-l2-13b": { "max_tokens": 8192, - "input_cost_per_token": 0.000001875, - "output_cost_per_token": 0.000001875, + "input_cost_per_token": 1.875e-06, + "output_cost_per_token": 1.875e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/jondurbin/airoboros-l2-70b-2.1": { "max_tokens": 4096, - "input_cost_per_token": 0.000013875, - "output_cost_per_token": 0.000013875, + "input_cost_per_token": 1.3875e-05, + "output_cost_per_token": 1.3875e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/undi95/remm-slerp-l2-13b": { "max_tokens": 6144, - "input_cost_per_token": 0.000001875, - "output_cost_per_token": 0.000001875, + "input_cost_per_token": 1.875e-06, + "output_cost_per_token": 1.875e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/pygmalionai/mythalion-13b": { "max_tokens": 4096, - "input_cost_per_token": 0.000001875, - "output_cost_per_token": 0.000001875, + "input_cost_per_token": 1.875e-06, + "output_cost_per_token": 1.875e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/mistralai/mistral-7b-instruct": { "max_tokens": 8192, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000013, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 1.3e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true @@ -8877,8 +9482,8 @@ "max_tokens": 33792, "max_input_tokens": 33792, "max_output_tokens": 33792, - "input_cost_per_token": 0.00000018, - "output_cost_per_token": 0.00000018, + "input_cost_per_token": 1.8e-07, + "output_cost_per_token": 1.8e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true @@ -8887,8 +9492,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 1.5e-05, "litellm_provider": "ai21", "mode": "completion" }, @@ -8896,8 +9501,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "ai21", "mode": "chat", "supports_tool_choice": true @@ -8906,8 +9511,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000008, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, "litellm_provider": "ai21", "mode": "chat", "supports_tool_choice": true @@ -8916,8 +9521,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "ai21", "mode": "chat", "supports_tool_choice": true @@ -8926,8 +9531,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "ai21", "mode": "chat", "supports_tool_choice": true @@ -8936,8 +9541,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000008, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, "litellm_provider": "ai21", "mode": "chat", "supports_tool_choice": true @@ -8946,8 +9551,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000008, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, "litellm_provider": "ai21", "mode": "chat", "supports_tool_choice": true @@ -8956,8 +9561,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "ai21", "mode": "chat", "supports_tool_choice": true @@ -8966,8 +9571,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 1e-05, "litellm_provider": "ai21", "mode": "completion" }, @@ -8975,8 +9580,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "ai21", "mode": "completion" }, @@ -8984,8 +9589,8 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "nlp_cloud", "mode": "completion" }, @@ -8993,68 +9598,68 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "nlp_cloud", "mode": "chat" }, "luminous-base": { - "max_tokens": 2048, - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.000033, + "max_tokens": 2048, + "input_cost_per_token": 3e-05, + "output_cost_per_token": 3.3e-05, "litellm_provider": "aleph_alpha", "mode": "completion" }, "luminous-base-control": { - "max_tokens": 2048, - "input_cost_per_token": 0.0000375, - "output_cost_per_token": 0.00004125, + "max_tokens": 2048, + "input_cost_per_token": 3.75e-05, + "output_cost_per_token": 4.125e-05, "litellm_provider": "aleph_alpha", "mode": "chat" }, "luminous-extended": { - "max_tokens": 2048, - "input_cost_per_token": 0.000045, - "output_cost_per_token": 0.0000495, + "max_tokens": 2048, + "input_cost_per_token": 4.5e-05, + "output_cost_per_token": 4.95e-05, "litellm_provider": "aleph_alpha", "mode": "completion" }, "luminous-extended-control": { - "max_tokens": 2048, - "input_cost_per_token": 0.00005625, - "output_cost_per_token": 0.000061875, + "max_tokens": 2048, + "input_cost_per_token": 5.625e-05, + "output_cost_per_token": 6.1875e-05, "litellm_provider": "aleph_alpha", "mode": "chat" }, "luminous-supreme": { - "max_tokens": 2048, + "max_tokens": 2048, "input_cost_per_token": 0.000175, "output_cost_per_token": 0.0001925, "litellm_provider": "aleph_alpha", "mode": "completion" }, "luminous-supreme-control": { - "max_tokens": 2048, + "max_tokens": 2048, "input_cost_per_token": 0.00021875, "output_cost_per_token": 0.000240625, "litellm_provider": "aleph_alpha", "mode": "chat" }, "ai21.j2-mid-v1": { - "max_tokens": 8191, - "max_input_tokens": 8191, - "max_output_tokens": 8191, - "input_cost_per_token": 0.0000125, - "output_cost_per_token": 0.0000125, + "max_tokens": 8191, + "max_input_tokens": 8191, + "max_output_tokens": 8191, + "input_cost_per_token": 1.25e-05, + "output_cost_per_token": 1.25e-05, "litellm_provider": "bedrock", "mode": "chat" }, "ai21.j2-ultra-v1": { - "max_tokens": 8191, - "max_input_tokens": 8191, - "max_output_tokens": 8191, - "input_cost_per_token": 0.0000188, - "output_cost_per_token": 0.0000188, + "max_tokens": 8191, + "max_input_tokens": 8191, + "max_output_tokens": 8191, + "input_cost_per_token": 1.88e-05, + "output_cost_per_token": 1.88e-05, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9062,8 +9667,8 @@ "max_tokens": 4096, "max_input_tokens": 70000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000007, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_system_messages": true @@ -9072,8 +9677,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000008, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9081,8 +9686,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9100,58 +9705,58 @@ "mode": "rerank" }, "amazon.titan-text-lite-v1": { - "max_tokens": 4000, + "max_tokens": 4000, "max_input_tokens": 42000, - "max_output_tokens": 4000, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000004, + "max_output_tokens": 4000, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "bedrock", "mode": "chat" }, "amazon.titan-text-express-v1": { - "max_tokens": 8000, + "max_tokens": 8000, "max_input_tokens": 42000, - "max_output_tokens": 8000, - "input_cost_per_token": 0.0000013, - "output_cost_per_token": 0.0000017, + "max_output_tokens": 8000, + "input_cost_per_token": 1.3e-06, + "output_cost_per_token": 1.7e-06, "litellm_provider": "bedrock", "mode": "chat" }, "amazon.titan-text-premier-v1:0": { - "max_tokens": 32000, + "max_tokens": 32000, "max_input_tokens": 42000, - "max_output_tokens": 32000, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "max_output_tokens": 32000, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "litellm_provider": "bedrock", "mode": "chat" }, "amazon.titan-embed-text-v1": { - "max_tokens": 8192, - "max_input_tokens": 8192, + "max_tokens": 8192, + "max_input_tokens": 8192, "output_vector_size": 1536, - "input_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0.0, - "litellm_provider": "bedrock", + "litellm_provider": "bedrock", "mode": "embedding" }, "amazon.titan-embed-text-v2:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, + "max_tokens": 8192, + "max_input_tokens": 8192, "output_vector_size": 1024, - "input_cost_per_token": 0.0000002, + "input_cost_per_token": 2e-07, "output_cost_per_token": 0.0, - "litellm_provider": "bedrock", + "litellm_provider": "bedrock", "mode": "embedding" }, "amazon.titan-embed-image-v1": { - "max_tokens": 128, - "max_input_tokens": 128, + "max_tokens": 128, + "max_input_tokens": 128, "output_vector_size": 1024, - "input_cost_per_token": 0.0000008, - "input_cost_per_image": 0.00006, + "input_cost_per_token": 8e-07, + "input_cost_per_image": 6e-05, "output_cost_per_token": 0.0, - "litellm_provider": "bedrock", + "litellm_provider": "bedrock", "supports_image_input": true, "supports_embedding_image_input": true, "mode": "embedding", @@ -9164,8 +9769,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9174,8 +9779,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000045, - "output_cost_per_token": 0.0000007, + "input_cost_per_token": 4.5e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9184,8 +9789,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9195,8 +9800,8 @@ "max_tokens": 8191, "max_input_tokens": 128000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000009, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 9e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9206,8 +9811,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9217,8 +9822,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000045, - "output_cost_per_token": 0.0000007, + "input_cost_per_token": 4.5e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9227,8 +9832,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000045, - "output_cost_per_token": 0.0000007, + "input_cost_per_token": 4.5e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9237,8 +9842,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000059, - "output_cost_per_token": 0.00000091, + "input_cost_per_token": 5.9e-07, + "output_cost_per_token": 9.1e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9247,8 +9852,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9257,8 +9862,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9267,8 +9872,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.00000026, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2.6e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9277,8 +9882,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9288,8 +9893,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9299,19 +9904,19 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000104, - "output_cost_per_token": 0.0000312, + "input_cost_per_token": 1.04e-05, + "output_cost_per_token": 3.12e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, "supports_tool_choice": true }, "amazon.nova-micro-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 300000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000035, - "output_cost_per_token": 0.00000014, + "input_cost_per_token": 3.5e-08, + "output_cost_per_token": 1.4e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9322,8 +9927,8 @@ "max_tokens": 4096, "max_input_tokens": 300000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000035, - "output_cost_per_token": 0.00000014, + "input_cost_per_token": 3.5e-08, + "output_cost_per_token": 1.4e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9331,11 +9936,11 @@ "supports_response_schema": true }, "eu.amazon.nova-micro-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 300000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000046, - "output_cost_per_token": 0.000000184, + "input_cost_per_token": 4.6e-08, + "output_cost_per_token": 1.84e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9343,11 +9948,11 @@ "supports_response_schema": true }, "amazon.nova-lite-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000006, - "output_cost_per_token": 0.00000024, + "input_cost_per_token": 6e-08, + "output_cost_per_token": 2.4e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9360,8 +9965,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000006, - "output_cost_per_token": 0.00000024, + "input_cost_per_token": 6e-08, + "output_cost_per_token": 2.4e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9371,11 +9976,11 @@ "supports_response_schema": true }, "eu.amazon.nova-lite-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000078, - "output_cost_per_token": 0.000000312, + "input_cost_per_token": 7.8e-08, + "output_cost_per_token": 3.12e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9385,11 +9990,11 @@ "supports_response_schema": true }, "amazon.nova-pro-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 300000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.0000032, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 3.2e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9402,8 +10007,8 @@ "max_tokens": 4096, "max_input_tokens": 300000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.0000032, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 3.2e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9413,17 +10018,17 @@ "supports_response_schema": true }, "1024-x-1024/50-steps/bedrock/amazon.nova-canvas-v1:0": { - "max_input_tokens": 2600, - "output_cost_per_image": 0.06, - "litellm_provider": "bedrock", - "mode": "image_generation" + "max_input_tokens": 2600, + "output_cost_per_image": 0.06, + "litellm_provider": "bedrock", + "mode": "image_generation" }, "eu.amazon.nova-pro-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 300000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000105, - "output_cost_per_token": 0.0000042, + "input_cost_per_token": 1.05e-06, + "output_cost_per_token": 4.2e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9437,8 +10042,8 @@ "max_tokens": 4096, "max_input_tokens": 1000000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.0000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1.25e-05, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9448,11 +10053,11 @@ "supports_response_schema": true }, "anthropic.claude-3-sonnet-20240229-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9462,11 +10067,11 @@ "supports_tool_choice": true }, "bedrock/invoke/anthropic.claude-3-5-sonnet-20240620-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9478,11 +10083,11 @@ } }, "anthropic.claude-3-5-sonnet-20240620-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9495,15 +10100,15 @@ "max_tokens": 32000, "max_input_tokens": 200000, "max_output_tokens": 32000, - "input_cost_per_token": 15e-6, - "output_cost_per_token": 75e-6, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 18.75e-6, - "cache_read_input_token_cost": 1.5e-6, + "cache_creation_input_token_cost": 1.875e-05, + "cache_read_input_token_cost": 1.5e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9521,15 +10126,15 @@ "max_tokens": 64000, "max_input_tokens": 200000, "max_output_tokens": 64000, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 15e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 3.75e-6, - "cache_read_input_token_cost": 0.3e-6, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9548,16 +10153,16 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, "supports_vision": true, "supports_assistant_prefill": true, - "supports_prompt_caching": true, + "supports_prompt_caching": true, "supports_response_schema": true, "supports_pdf_input": true, "supports_reasoning": true, @@ -9568,26 +10173,26 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, "supports_vision": true, "supports_pdf_input": true, "supports_assistant_prefill": true, - "supports_prompt_caching": true, + "supports_prompt_caching": true, "supports_response_schema": true, "supports_tool_choice": true }, "anthropic.claude-3-haiku-20240307-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9600,10 +10205,10 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.000004, - "cache_creation_input_token_cost": 0.000001, - "cache_read_input_token_cost": 0.00000008, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 4e-06, + "cache_creation_input_token_cost": 1e-06, + "cache_read_input_token_cost": 8e-08, "litellm_provider": "bedrock", "mode": "chat", "supports_assistant_prefill": true, @@ -9617,8 +10222,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9630,8 +10235,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9644,8 +10249,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9659,10 +10264,10 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9678,16 +10283,16 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, "supports_vision": true, "supports_assistant_prefill": true, - "supports_prompt_caching": true, + "supports_prompt_caching": true, "supports_response_schema": true, "supports_pdf_input": true, "supports_tool_choice": true, @@ -9697,15 +10302,15 @@ "max_tokens": 32000, "max_input_tokens": 200000, "max_output_tokens": 32000, - "input_cost_per_token": 15e-6, - "output_cost_per_token": 75e-6, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 18.75e-6, - "cache_read_input_token_cost": 1.5e-6, + "cache_creation_input_token_cost": 1.875e-05, + "cache_read_input_token_cost": 1.5e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9723,15 +10328,15 @@ "max_tokens": 64000, "max_input_tokens": 200000, "max_output_tokens": 64000, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 15e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 3.75e-6, - "cache_read_input_token_cost": 0.3e-6, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9749,8 +10354,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9763,10 +10368,10 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.000004, - "cache_creation_input_token_cost": 0.000001, - "cache_read_input_token_cost": 0.00000008, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 4e-06, + "cache_creation_input_token_cost": 1e-06, + "cache_read_input_token_cost": 8e-08, "litellm_provider": "bedrock", "mode": "chat", "supports_assistant_prefill": true, @@ -9780,8 +10385,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9793,8 +10398,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9807,8 +10412,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9822,8 +10427,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9839,14 +10444,14 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, "supports_vision": true, "supports_assistant_prefill": true, - "supports_prompt_caching": true, + "supports_prompt_caching": true, "supports_response_schema": true, "supports_pdf_input": true, "supports_tool_choice": true, @@ -9856,8 +10461,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9870,15 +10475,15 @@ "max_tokens": 32000, "max_input_tokens": 200000, "max_output_tokens": 32000, - "input_cost_per_token": 15e-6, - "output_cost_per_token": 75e-6, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 18.75e-6, - "cache_read_input_token_cost": 1.5e-6, + "cache_creation_input_token_cost": 1.875e-05, + "cache_read_input_token_cost": 1.5e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9896,15 +10501,15 @@ "max_tokens": 64000, "max_input_tokens": 200000, "max_output_tokens": 64000, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 15e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 3.75e-6, - "cache_read_input_token_cost": 0.3e-6, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9922,8 +10527,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9937,8 +10542,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9947,46 +10552,46 @@ "supports_tool_choice": true }, "anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-east-1/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/us-west-2/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/ap-northeast-1/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/ap-northeast-1/1-month-commitment/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.0455, @@ -9995,7 +10600,7 @@ "mode": "chat" }, "bedrock/ap-northeast-1/6-month-commitment/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.02527, @@ -10004,112 +10609,112 @@ "mode": "chat" }, "bedrock/eu-central-1/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "max_output_tokens": 8191, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/eu-central-1/1-month-commitment/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.0415, "output_cost_per_second": 0.0415, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/eu-central-1/6-month-commitment/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.02305, "output_cost_per_second": 0.02305, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-east-1/1-month-commitment/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.0175, "output_cost_per_second": 0.0175, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-east-1/6-month-commitment/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.00972, "output_cost_per_second": 0.00972, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-west-2/1-month-commitment/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.0175, "output_cost_per_second": 0.0175, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-west-2/6-month-commitment/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.00972, "output_cost_per_second": 0.00972, "litellm_provider": "bedrock", "mode": "chat" }, "anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "max_output_tokens": 8191, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/us-east-1/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "max_output_tokens": 8191, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/us-west-2/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "max_output_tokens": 8191, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/ap-northeast-1/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "max_output_tokens": 8191, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/ap-northeast-1/1-month-commitment/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.0455, "output_cost_per_second": 0.0455, "litellm_provider": "bedrock", @@ -10117,9 +10722,9 @@ "supports_tool_choice": true }, "bedrock/ap-northeast-1/6-month-commitment/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.02527, "output_cost_per_second": 0.02527, "litellm_provider": "bedrock", @@ -10127,19 +10732,19 @@ "supports_tool_choice": true }, "bedrock/eu-central-1/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "max_output_tokens": 8191, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/eu-central-1/1-month-commitment/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.0415, "output_cost_per_second": 0.0415, "litellm_provider": "bedrock", @@ -10147,9 +10752,9 @@ "supports_tool_choice": true }, "bedrock/eu-central-1/6-month-commitment/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.02305, "output_cost_per_second": 0.02305, "litellm_provider": "bedrock", @@ -10157,9 +10762,9 @@ "supports_tool_choice": true }, "bedrock/us-east-1/1-month-commitment/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.0175, "output_cost_per_second": 0.0175, "litellm_provider": "bedrock", @@ -10167,9 +10772,9 @@ "supports_tool_choice": true }, "bedrock/us-east-1/6-month-commitment/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.00972, "output_cost_per_second": 0.00972, "litellm_provider": "bedrock", @@ -10177,9 +10782,9 @@ "supports_tool_choice": true }, "bedrock/us-west-2/1-month-commitment/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.0175, "output_cost_per_second": 0.0175, "litellm_provider": "bedrock", @@ -10187,9 +10792,9 @@ "supports_tool_choice": true }, "bedrock/us-west-2/6-month-commitment/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.00972, "output_cost_per_second": 0.00972, "litellm_provider": "bedrock", @@ -10197,48 +10802,48 @@ "supports_tool_choice": true }, "anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/us-east-1/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/us-west-2/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/ap-northeast-1/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/ap-northeast-1/1-month-commitment/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.0455, "output_cost_per_second": 0.0455, @@ -10247,8 +10852,8 @@ "supports_tool_choice": true }, "bedrock/ap-northeast-1/6-month-commitment/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.02527, "output_cost_per_second": 0.02527, @@ -10257,18 +10862,18 @@ "supports_tool_choice": true }, "bedrock/eu-central-1/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/eu-central-1/1-month-commitment/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.0415, "output_cost_per_second": 0.0415, @@ -10277,8 +10882,8 @@ "supports_tool_choice": true }, "bedrock/eu-central-1/6-month-commitment/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.02305, "output_cost_per_second": 0.02305, @@ -10287,8 +10892,8 @@ "supports_tool_choice": true }, "bedrock/us-east-1/1-month-commitment/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.0175, "output_cost_per_second": 0.0175, @@ -10297,8 +10902,8 @@ "supports_tool_choice": true }, "bedrock/us-east-1/6-month-commitment/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.00972, "output_cost_per_second": 0.00972, @@ -10307,8 +10912,8 @@ "supports_tool_choice": true }, "bedrock/us-west-2/1-month-commitment/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.0175, "output_cost_per_second": 0.0175, @@ -10317,8 +10922,8 @@ "supports_tool_choice": true }, "bedrock/us-west-2/6-month-commitment/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.00972, "output_cost_per_second": 0.00972, @@ -10327,28 +10932,28 @@ "supports_tool_choice": true }, "anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.0000024, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 2.4e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/us-east-1/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.0000024, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 2.4e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/us-east-1/1-month-commitment/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.011, "output_cost_per_second": 0.011, @@ -10357,8 +10962,8 @@ "supports_tool_choice": true }, "bedrock/us-east-1/6-month-commitment/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.00611, "output_cost_per_second": 0.00611, @@ -10367,8 +10972,8 @@ "supports_tool_choice": true }, "bedrock/us-west-2/1-month-commitment/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.011, "output_cost_per_second": 0.011, @@ -10377,8 +10982,8 @@ "supports_tool_choice": true }, "bedrock/us-west-2/6-month-commitment/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.00611, "output_cost_per_second": 0.00611, @@ -10387,28 +10992,28 @@ "supports_tool_choice": true }, "bedrock/us-west-2/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.0000024, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 2.4e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/ap-northeast-1/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000223, - "output_cost_per_token": 0.00000755, + "input_cost_per_token": 2.23e-06, + "output_cost_per_token": 7.55e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/ap-northeast-1/1-month-commitment/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.01475, "output_cost_per_second": 0.01475, @@ -10417,8 +11022,8 @@ "supports_tool_choice": true }, "bedrock/ap-northeast-1/6-month-commitment/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.008194, "output_cost_per_second": 0.008194, @@ -10427,18 +11032,18 @@ "supports_tool_choice": true }, "bedrock/eu-central-1/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000248, - "output_cost_per_token": 0.00000838, + "input_cost_per_token": 2.48e-06, + "output_cost_per_token": 8.38e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/eu-central-1/1-month-commitment/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.01635, "output_cost_per_second": 0.01635, @@ -10447,8 +11052,8 @@ "supports_tool_choice": true }, "bedrock/eu-central-1/6-month-commitment/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.009083, "output_cost_per_second": 0.009083, @@ -10470,19 +11075,19 @@ "mode": "rerank" }, "cohere.command-text-v14": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.0000020, + "max_output_tokens": 4096, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/*/1-month-commitment/cohere.command-text-v14": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_second": 0.011, "output_cost_per_second": 0.011, "litellm_provider": "bedrock", @@ -10490,9 +11095,9 @@ "supports_tool_choice": true }, "bedrock/*/6-month-commitment/cohere.command-text-v14": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_second": 0.0066027, "output_cost_per_second": 0.0066027, "litellm_provider": "bedrock", @@ -10500,19 +11105,19 @@ "supports_tool_choice": true }, "cohere.command-light-text-v14": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000006, + "max_output_tokens": 4096, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/*/1-month-commitment/cohere.command-light-text-v14": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_second": 0.001902, "output_cost_per_second": 0.001902, "litellm_provider": "bedrock", @@ -10520,9 +11125,9 @@ "supports_tool_choice": true }, "bedrock/*/6-month-commitment/cohere.command-light-text-v14": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_second": 0.0011416, "output_cost_per_second": 0.0011416, "litellm_provider": "bedrock", @@ -10530,226 +11135,225 @@ "supports_tool_choice": true }, "cohere.command-r-plus-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000030, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "cohere.command-r-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "cohere.embed-english-v3": { - "max_tokens": 512, - "max_input_tokens": 512, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, - "litellm_provider": "bedrock", + "max_tokens": 512, + "max_input_tokens": 512, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, + "litellm_provider": "bedrock", "mode": "embedding", "supports_embedding_image_input": true }, "cohere.embed-multilingual-v3": { - "max_tokens": 512, - "max_input_tokens": 512, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, + "max_tokens": 512, + "max_input_tokens": 512, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "bedrock", "mode": "embedding", "supports_embedding_image_input": true }, "us.deepseek.r1-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000135, - "output_cost_per_token": 0.0000054, + "input_cost_per_token": 1.35e-06, + "output_cost_per_token": 5.4e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_reasoning": true, - "supports_function_calling": false, + "supports_function_calling": false, "supports_tool_choice": false - }, "meta.llama3-3-70b-instruct-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000072, - "output_cost_per_token": 0.00000072, + "input_cost_per_token": 7.2e-07, + "output_cost_per_token": 7.2e-07, "litellm_provider": "bedrock_converse", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "meta.llama2-13b-chat-v1": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000075, - "output_cost_per_token": 0.000001, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 7.5e-07, + "output_cost_per_token": 1e-06, "litellm_provider": "bedrock", "mode": "chat" }, "meta.llama2-70b-chat-v1": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000195, - "output_cost_per_token": 0.00000256, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 1.95e-06, + "output_cost_per_token": 2.56e-06, "litellm_provider": "bedrock", "mode": "chat" }, "meta.llama3-8b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000006, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-east-1/meta.llama3-8b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000006, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-west-1/meta.llama3-8b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000006, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/ap-south-1/meta.llama3-8b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000036, - "output_cost_per_token": 0.00000072, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3.6e-07, + "output_cost_per_token": 7.2e-07, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/ca-central-1/meta.llama3-8b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000035, - "output_cost_per_token": 0.00000069, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 6.9e-07, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/eu-west-1/meta.llama3-8b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000032, - "output_cost_per_token": 0.00000065, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3.2e-07, + "output_cost_per_token": 6.5e-07, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/eu-west-2/meta.llama3-8b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000039, - "output_cost_per_token": 0.00000078, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3.9e-07, + "output_cost_per_token": 7.8e-07, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/sa-east-1/meta.llama3-8b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.00000101, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.01e-06, "litellm_provider": "bedrock", "mode": "chat" }, "meta.llama3-70b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000265, - "output_cost_per_token": 0.0000035, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 2.65e-06, + "output_cost_per_token": 3.5e-06, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-east-1/meta.llama3-70b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000265, - "output_cost_per_token": 0.0000035, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 2.65e-06, + "output_cost_per_token": 3.5e-06, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-west-1/meta.llama3-70b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000265, - "output_cost_per_token": 0.0000035, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 2.65e-06, + "output_cost_per_token": 3.5e-06, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/ap-south-1/meta.llama3-70b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000318, - "output_cost_per_token": 0.0000042, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3.18e-06, + "output_cost_per_token": 4.2e-06, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/ca-central-1/meta.llama3-70b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000305, - "output_cost_per_token": 0.00000403, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3.05e-06, + "output_cost_per_token": 4.03e-06, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/eu-west-1/meta.llama3-70b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000286, - "output_cost_per_token": 0.00000378, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 2.86e-06, + "output_cost_per_token": 3.78e-06, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/eu-west-2/meta.llama3-70b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000345, - "output_cost_per_token": 0.00000455, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3.45e-06, + "output_cost_per_token": 4.55e-06, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/sa-east-1/meta.llama3-70b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000445, - "output_cost_per_token": 0.00000588, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 4.45e-06, + "output_cost_per_token": 5.88e-06, "litellm_provider": "bedrock", "mode": "chat" }, @@ -10757,143 +11361,143 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000022, - "output_cost_per_token": 0.00000022, + "input_cost_per_token": 2.2e-07, + "output_cost_per_token": 2.2e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "us.meta.llama3-1-8b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000022, - "output_cost_per_token": 0.00000022, + "input_cost_per_token": 2.2e-07, + "output_cost_per_token": 2.2e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "meta.llama3-1-70b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000099, - "output_cost_per_token": 0.00000099, + "input_cost_per_token": 9.9e-07, + "output_cost_per_token": 9.9e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "us.meta.llama3-1-70b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000099, - "output_cost_per_token": 0.00000099, + "input_cost_per_token": 9.9e-07, + "output_cost_per_token": 9.9e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "meta.llama3-1-405b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000532, - "output_cost_per_token": 0.000016, + "input_cost_per_token": 5.32e-06, + "output_cost_per_token": 1.6e-05, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "us.meta.llama3-1-405b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000532, - "output_cost_per_token": 0.000016, + "input_cost_per_token": 5.32e-06, + "output_cost_per_token": 1.6e-05, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "meta.llama3-2-1b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 1e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "us.meta.llama3-2-1b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 1e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "eu.meta.llama3-2-1b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000013, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 1.3e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "meta.llama3-2-3b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "us.meta.llama3-2-3b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "eu.meta.llama3-2-3b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000019, - "output_cost_per_token": 0.00000019, + "input_cost_per_token": 1.9e-07, + "output_cost_per_token": 1.9e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "meta.llama3-2-11b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000035, - "output_cost_per_token": 0.00000035, + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 3.5e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false, "supports_vision": true }, @@ -10901,11 +11505,11 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000035, - "output_cost_per_token": 0.00000035, + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 3.5e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false, "supports_vision": true }, @@ -10913,11 +11517,11 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false, "supports_vision": true }, @@ -10925,130 +11529,154 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false, "supports_vision": true }, "us.meta.llama3-3-70b-instruct-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000072, - "output_cost_per_token": 0.00000072, + "input_cost_per_token": 7.2e-07, + "output_cost_per_token": 7.2e-07, "litellm_provider": "bedrock_converse", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "meta.llama4-maverick-17b-instruct-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00024e-3, - "input_cost_per_token_batches": 0.00012e-3, - "output_cost_per_token": 0.00097e-3, - "output_cost_per_token_batches": 0.000485e-3, + "input_cost_per_token": 2.4e-07, + "input_cost_per_token_batches": 1.2e-07, + "output_cost_per_token": 9.7e-07, + "output_cost_per_token_batches": 4.85e-07, "litellm_provider": "bedrock_converse", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text", "code"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] }, "us.meta.llama4-maverick-17b-instruct-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00024e-3, - "input_cost_per_token_batches": 0.00012e-3, - "output_cost_per_token": 0.00097e-3, - "output_cost_per_token_batches": 0.000485e-3, + "input_cost_per_token": 2.4e-07, + "input_cost_per_token_batches": 1.2e-07, + "output_cost_per_token": 9.7e-07, + "output_cost_per_token_batches": 4.85e-07, "litellm_provider": "bedrock_converse", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text", "code"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] }, "meta.llama4-scout-17b-instruct-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00017e-3, - "input_cost_per_token_batches": 0.000085e-3, - "output_cost_per_token": 0.00066e-3, - "output_cost_per_token_batches": 0.00033e-3, + "input_cost_per_token": 1.7e-07, + "input_cost_per_token_batches": 8.5e-08, + "output_cost_per_token": 6.6e-07, + "output_cost_per_token_batches": 3.3e-07, "litellm_provider": "bedrock_converse", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text", "code"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] }, "us.meta.llama4-scout-17b-instruct-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00017e-3, - "input_cost_per_token_batches": 0.000085e-3, - "output_cost_per_token": 0.00066e-3, - "output_cost_per_token_batches": 0.00033e-3, + "input_cost_per_token": 1.7e-07, + "input_cost_per_token_batches": 8.5e-08, + "output_cost_per_token": 6.6e-07, + "output_cost_per_token_batches": 3.3e-07, "litellm_provider": "bedrock_converse", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text", "code"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] }, "512-x-512/50-steps/stability.stable-diffusion-xl-v0": { - "max_tokens": 77, - "max_input_tokens": 77, + "max_tokens": 77, + "max_input_tokens": 77, "output_cost_per_image": 0.018, "litellm_provider": "bedrock", "mode": "image_generation" }, "512-x-512/max-steps/stability.stable-diffusion-xl-v0": { - "max_tokens": 77, - "max_input_tokens": 77, + "max_tokens": 77, + "max_input_tokens": 77, "output_cost_per_image": 0.036, "litellm_provider": "bedrock", "mode": "image_generation" }, "max-x-max/50-steps/stability.stable-diffusion-xl-v0": { - "max_tokens": 77, - "max_input_tokens": 77, + "max_tokens": 77, + "max_input_tokens": 77, "output_cost_per_image": 0.036, "litellm_provider": "bedrock", "mode": "image_generation" }, "max-x-max/max-steps/stability.stable-diffusion-xl-v0": { - "max_tokens": 77, - "max_input_tokens": 77, + "max_tokens": 77, + "max_input_tokens": 77, "output_cost_per_image": 0.072, "litellm_provider": "bedrock", "mode": "image_generation" }, "1024-x-1024/50-steps/stability.stable-diffusion-xl-v1": { - "max_tokens": 77, - "max_input_tokens": 77, + "max_tokens": 77, + "max_input_tokens": 77, "output_cost_per_image": 0.04, "litellm_provider": "bedrock", "mode": "image_generation" }, "1024-x-1024/max-steps/stability.stable-diffusion-xl-v1": { - "max_tokens": 77, - "max_input_tokens": 77, + "max_tokens": 77, + "max_input_tokens": 77, "output_cost_per_image": 0.08, "litellm_provider": "bedrock", "mode": "image_generation" }, "stability.sd3-large-v1:0": { - "max_tokens": 77, - "max_input_tokens": 77, + "max_tokens": 77, + "max_input_tokens": 77, "output_cost_per_image": 0.08, "litellm_provider": "bedrock", "mode": "image_generation" @@ -11075,8 +11703,8 @@ "mode": "image_generation" }, "stability.stable-image-ultra-v1:0": { - "max_tokens": 77, - "max_input_tokens": 77, + "max_tokens": 77, + "max_input_tokens": 77, "output_cost_per_image": 0.14, "litellm_provider": "bedrock", "mode": "image_generation" @@ -11089,111 +11717,111 @@ "mode": "image_generation" }, "sagemaker/meta-textgeneration-llama-2-7b": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "completion" }, "sagemaker/meta-textgeneration-llama-2-7b-f": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "chat" }, "sagemaker/meta-textgeneration-llama-2-13b": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "completion" }, "sagemaker/meta-textgeneration-llama-2-13b-f": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "chat" }, "sagemaker/meta-textgeneration-llama-2-70b": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "completion" }, "sagemaker/meta-textgeneration-llama-2-70b-b-f": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "chat" }, "together-ai-up-to-4b": { - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 1e-07, "litellm_provider": "together_ai", "mode": "chat" }, "together-ai-4.1b-8b": { - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "together_ai", "mode": "chat" }, "together-ai-8.1b-21b": { "max_tokens": 1000, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000003, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 3e-07, "litellm_provider": "together_ai", "mode": "chat" }, "together-ai-21.1b-41b": { - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.0000008, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 8e-07, "litellm_provider": "together_ai", "mode": "chat" }, "together-ai-41.1b-80b": { - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "together_ai", "mode": "chat" }, "together-ai-81.1b-110b": { - "input_cost_per_token": 0.0000018, - "output_cost_per_token": 0.0000018, + "input_cost_per_token": 1.8e-06, + "output_cost_per_token": 1.8e-06, "litellm_provider": "together_ai", "mode": "chat" }, "together-ai-embedding-up-to-150m": { - "input_cost_per_token": 0.000000008, + "input_cost_per_token": 8e-09, "output_cost_per_token": 0.0, "litellm_provider": "together_ai", "mode": "embedding" }, "together-ai-embedding-151m-to-350m": { - "input_cost_per_token": 0.000000016, + "input_cost_per_token": 1.6e-08, "output_cost_per_token": 0.0, "litellm_provider": "together_ai", "mode": "embedding" }, "together_ai/meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo": { - "input_cost_per_token": 0.00000018, - "output_cost_per_token": 0.00000018, + "input_cost_per_token": 1.8e-07, + "output_cost_per_token": 1.8e-07, "litellm_provider": "together_ai", "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -11202,8 +11830,8 @@ "supports_tool_choice": true }, "together_ai/meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo": { - "input_cost_per_token": 0.00000088, - "output_cost_per_token": 0.00000088, + "input_cost_per_token": 8.8e-07, + "output_cost_per_token": 8.8e-07, "litellm_provider": "together_ai", "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -11212,8 +11840,8 @@ "supports_tool_choice": true }, "together_ai/meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": { - "input_cost_per_token": 0.0000035, - "output_cost_per_token": 0.0000035, + "input_cost_per_token": 3.5e-06, + "output_cost_per_token": 3.5e-06, "litellm_provider": "together_ai", "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -11221,8 +11849,8 @@ "supports_tool_choice": true }, "together_ai/meta-llama/Llama-3.3-70B-Instruct-Turbo": { - "input_cost_per_token": 0.00000088, - "output_cost_per_token": 0.00000088, + "input_cost_per_token": 8.8e-07, + "output_cost_per_token": 8.8e-07, "litellm_provider": "together_ai", "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -11241,8 +11869,8 @@ "supports_tool_choice": true }, "together_ai/mistralai/Mixtral-8x7B-Instruct-v0.1": { - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "together_ai", "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -11315,9 +11943,9 @@ "supports_tool_choice": true }, "ollama/codegemma": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", @@ -11330,7 +11958,7 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "chat", + "mode": "chat", "supports_function_calling": false }, "ollama/deepseek-coder-v2-instruct": { @@ -11340,7 +11968,7 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "chat", + "mode": "chat", "supports_function_calling": true }, "ollama/deepseek-coder-v2-base": { @@ -11350,7 +11978,7 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "completion", + "mode": "completion", "supports_function_calling": true }, "ollama/deepseek-coder-v2-lite-instruct": { @@ -11360,7 +11988,7 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "chat", + "mode": "chat", "supports_function_calling": true }, "ollama/deepseek-coder-v2-lite-base": { @@ -11370,7 +11998,7 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "completion", + "mode": "completion", "supports_function_calling": true }, "ollama/internlm2_5-20b-chat": { @@ -11380,49 +12008,49 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "chat", + "mode": "chat", "supports_function_calling": true }, "ollama/llama2": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", "mode": "chat" }, "ollama/llama2:7b": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", "mode": "chat" }, "ollama/llama2:13b": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", "mode": "chat" }, "ollama/llama2:70b": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", "mode": "chat" }, "ollama/llama2-uncensored": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", @@ -11462,7 +12090,7 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "chat", + "mode": "chat", "supports_function_calling": true }, "ollama/mistral-large-instruct-2407": { @@ -11526,8 +12154,8 @@ "supports_function_calling": true }, "ollama/codellama": { - "max_tokens": 4096, - "max_input_tokens": 4096, + "max_tokens": 4096, + "max_input_tokens": 4096, "max_output_tokens": 4096, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, @@ -11535,8 +12163,8 @@ "mode": "completion" }, "ollama/orca-mini": { - "max_tokens": 4096, - "max_input_tokens": 4096, + "max_tokens": 4096, + "max_input_tokens": 4096, "max_output_tokens": 4096, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, @@ -11556,8 +12184,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000090, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11566,8 +12194,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000022, - "output_cost_per_token": 0.00000022, + "input_cost_per_token": 2.2e-07, + "output_cost_per_token": 2.2e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11576,8 +12204,8 @@ "max_tokens": 8191, "max_input_tokens": 32768, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000013, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 1.3e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11586,8 +12214,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000090, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11596,8 +12224,8 @@ "max_tokens": 8191, "max_input_tokens": 32768, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000027, - "output_cost_per_token": 0.00000027, + "input_cost_per_token": 2.7e-07, + "output_cost_per_token": 2.7e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11606,8 +12234,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000060, - "output_cost_per_token": 0.00000060, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11616,8 +12244,8 @@ "max_tokens": 4096, "max_input_tokens": 32000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000027, - "output_cost_per_token": 0.00000027, + "input_cost_per_token": 2.7e-07, + "output_cost_per_token": 2.7e-07, "litellm_provider": "deepinfra", "mode": "completion" }, @@ -11625,8 +12253,8 @@ "max_tokens": 4096, "max_input_tokens": 16384, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000060, - "output_cost_per_token": 0.00000060, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11635,8 +12263,8 @@ "max_tokens": 8191, "max_input_tokens": 32768, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000027, - "output_cost_per_token": 0.00000027, + "input_cost_per_token": 2.7e-07, + "output_cost_per_token": 2.7e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11645,8 +12273,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000090, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11655,8 +12283,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000060, - "output_cost_per_token": 0.00000060, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11665,8 +12293,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000013, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 1.3e-07, "litellm_provider": "deepinfra", "mode": "completion" }, @@ -11674,8 +12302,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000090, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11684,8 +12312,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000022, - "output_cost_per_token": 0.00000022, + "input_cost_per_token": 2.2e-07, + "output_cost_per_token": 2.2e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11694,8 +12322,8 @@ "max_tokens": 8191, "max_input_tokens": 32768, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000020, - "output_cost_per_token": 0.00000020, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11704,8 +12332,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000013, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 1.3e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11714,8 +12342,8 @@ "max_tokens": 8191, "max_input_tokens": 8191, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000008, - "output_cost_per_token": 0.00000008, + "input_cost_per_token": 8e-08, + "output_cost_per_token": 8e-08, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11724,8 +12352,8 @@ "max_tokens": 8191, "max_input_tokens": 8191, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000059, - "output_cost_per_token": 0.00000079, + "input_cost_per_token": 5.9e-07, + "output_cost_per_token": 7.9e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11734,8 +12362,8 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 32768, - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_function_calling": true, @@ -11746,8 +12374,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000060, - "output_cost_per_token": 0.00000060, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "deepinfra", "mode": "completion" }, @@ -11755,160 +12383,160 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000013, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 1.3e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true }, - "perplexity/codellama-34b-instruct": { + "perplexity/codellama-34b-instruct": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000035, - "output_cost_per_token": 0.00000140, - "litellm_provider": "perplexity", - "mode": "chat" + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 1.4e-06, + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/codellama-70b-instruct": { + "perplexity/codellama-70b-instruct": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000280, - "litellm_provider": "perplexity", - "mode": "chat" + "input_cost_per_token": 7e-07, + "output_cost_per_token": 2.8e-06, + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/llama-3.1-70b-instruct": { + "perplexity/llama-3.1-70b-instruct": { "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000001, - "litellm_provider": "perplexity", - "mode": "chat" + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/llama-3.1-8b-instruct": { + "perplexity/llama-3.1-8b-instruct": { "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, - "litellm_provider": "perplexity", - "mode": "chat" + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/llama-3.1-sonar-huge-128k-online": { + "perplexity/llama-3.1-sonar-huge-128k-online": { "max_tokens": 127072, "max_input_tokens": 127072, "max_output_tokens": 127072, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000005, - "litellm_provider": "perplexity", + "input_cost_per_token": 5e-06, + "output_cost_per_token": 5e-06, + "litellm_provider": "perplexity", "mode": "chat", "deprecation_date": "2025-02-22" }, - "perplexity/llama-3.1-sonar-large-128k-online": { + "perplexity/llama-3.1-sonar-large-128k-online": { "max_tokens": 127072, "max_input_tokens": 127072, "max_output_tokens": 127072, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000001, - "litellm_provider": "perplexity", + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, + "litellm_provider": "perplexity", "mode": "chat", "deprecation_date": "2025-02-22" }, - "perplexity/llama-3.1-sonar-large-128k-chat": { + "perplexity/llama-3.1-sonar-large-128k-chat": { "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000001, - "litellm_provider": "perplexity", + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, + "litellm_provider": "perplexity", "mode": "chat", "deprecation_date": "2025-02-22" }, - "perplexity/llama-3.1-sonar-small-128k-chat": { + "perplexity/llama-3.1-sonar-small-128k-chat": { "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, - "litellm_provider": "perplexity", + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, + "litellm_provider": "perplexity", "mode": "chat", "deprecation_date": "2025-02-22" }, - "perplexity/llama-3.1-sonar-small-128k-online": { + "perplexity/llama-3.1-sonar-small-128k-online": { "max_tokens": 127072, "max_input_tokens": 127072, "max_output_tokens": 127072, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, - "litellm_provider": "perplexity", - "mode": "chat" , + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, + "litellm_provider": "perplexity", + "mode": "chat", "deprecation_date": "2025-02-22" }, - "perplexity/pplx-7b-chat": { + "perplexity/pplx-7b-chat": { "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000007, - "output_cost_per_token": 0.00000028, - "litellm_provider": "perplexity", - "mode": "chat" + "input_cost_per_token": 7e-08, + "output_cost_per_token": 2.8e-07, + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/pplx-70b-chat": { + "perplexity/pplx-70b-chat": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000280, - "litellm_provider": "perplexity", - "mode": "chat" + "input_cost_per_token": 7e-07, + "output_cost_per_token": 2.8e-06, + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/pplx-7b-online": { + "perplexity/pplx-7b-online": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000000, - "output_cost_per_token": 0.00000028, + "input_cost_per_token": 0.0, + "output_cost_per_token": 2.8e-07, "input_cost_per_request": 0.005, - "litellm_provider": "perplexity", - "mode": "chat" + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/pplx-70b-online": { + "perplexity/pplx-70b-online": { "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.0000000, - "output_cost_per_token": 0.00000280, + "max_output_tokens": 4096, + "input_cost_per_token": 0.0, + "output_cost_per_token": 2.8e-06, "input_cost_per_request": 0.005, - "litellm_provider": "perplexity", - "mode": "chat" + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/llama-2-70b-chat": { + "perplexity/llama-2-70b-chat": { "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000280, - "litellm_provider": "perplexity", - "mode": "chat" + "max_output_tokens": 4096, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 2.8e-06, + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/mistral-7b-instruct": { + "perplexity/mistral-7b-instruct": { "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000007, - "output_cost_per_token": 0.00000028, - "litellm_provider": "perplexity", - "mode": "chat" + "max_output_tokens": 4096, + "input_cost_per_token": 7e-08, + "output_cost_per_token": 2.8e-07, + "litellm_provider": "perplexity", + "mode": "chat" }, "perplexity/mixtral-8x7b-instruct": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000007, - "output_cost_per_token": 0.00000028, + "input_cost_per_token": 7e-08, + "output_cost_per_token": 2.8e-07, "litellm_provider": "perplexity", "mode": "chat" }, @@ -11916,8 +12544,8 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000007, - "output_cost_per_token": 0.00000028, + "input_cost_per_token": 7e-08, + "output_cost_per_token": 2.8e-07, "litellm_provider": "perplexity", "mode": "chat" }, @@ -11926,7 +12554,7 @@ "max_input_tokens": 12000, "max_output_tokens": 12000, "input_cost_per_token": 0, - "output_cost_per_token": 0.00000028, + "output_cost_per_token": 2.8e-07, "input_cost_per_request": 0.005, "litellm_provider": "perplexity", "mode": "chat" @@ -11935,8 +12563,8 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.0000018, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 1.8e-06, "litellm_provider": "perplexity", "mode": "chat" }, @@ -11945,7 +12573,7 @@ "max_input_tokens": 12000, "max_output_tokens": 12000, "input_cost_per_token": 0, - "output_cost_per_token": 0.0000018, + "output_cost_per_token": 1.8e-06, "input_cost_per_request": 0.005, "litellm_provider": "perplexity", "mode": "chat" @@ -11953,14 +12581,14 @@ "perplexity/sonar": { "max_tokens": 128000, "max_input_tokens": 128000, - "input_cost_per_token": 1e-6, - "output_cost_per_token": 1e-6, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, "litellm_provider": "perplexity", "mode": "chat", "search_context_cost_per_query": { - "search_context_size_low": 5e-3, - "search_context_size_medium": 8e-3, - "search_context_size_high": 12e-3 + "search_context_size_low": 0.005, + "search_context_size_medium": 0.008, + "search_context_size_high": 0.012 }, "supports_web_search": true }, @@ -11968,28 +12596,28 @@ "max_tokens": 8000, "max_input_tokens": 200000, "max_output_tokens": 8000, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 15e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "perplexity", "mode": "chat", "search_context_cost_per_query": { - "search_context_size_low": 6e-3, - "search_context_size_medium": 10e-3, - "search_context_size_high": 14e-3 + "search_context_size_low": 0.006, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.014 }, "supports_web_search": true }, "perplexity/sonar-reasoning": { "max_tokens": 128000, "max_input_tokens": 128000, - "input_cost_per_token": 1e-6, - "output_cost_per_token": 5e-6, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 5e-06, "litellm_provider": "perplexity", "mode": "chat", "search_context_cost_per_query": { - "search_context_size_low": 5e-3, - "search_context_size_medium": 8e-3, - "search_context_size_high": 14e-3 + "search_context_size_low": 0.005, + "search_context_size_medium": 0.008, + "search_context_size_high": 0.014 }, "supports_web_search": true, "supports_reasoning": true @@ -11997,14 +12625,14 @@ "perplexity/sonar-reasoning-pro": { "max_tokens": 128000, "max_input_tokens": 128000, - "input_cost_per_token": 2e-6, - "output_cost_per_token": 8e-6, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, "litellm_provider": "perplexity", "mode": "chat", "search_context_cost_per_query": { - "search_context_size_low": 6e-3, - "search_context_size_medium": 10e-3, - "search_context_size_high": 14e-3 + "search_context_size_low": 0.006, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.014 }, "supports_web_search": true, "supports_reasoning": true @@ -12012,15 +12640,15 @@ "perplexity/sonar-deep-research": { "max_tokens": 128000, "max_input_tokens": 128000, - "input_cost_per_token": 2e-6, - "output_cost_per_token": 8e-6, - "output_cost_per_reasoning_token": 3e-6, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, + "output_cost_per_reasoning_token": 3e-06, "litellm_provider": "perplexity", "mode": "chat", "search_context_cost_per_query": { - "search_context_size_low": 5e-3, - "search_context_size_medium": 5e-3, - "search_context_size_high": 5e-3 + "search_context_size_low": 0.005, + "search_context_size_medium": 0.005, + "search_context_size_high": 0.005 }, "supports_reasoning": true, "supports_web_search": true @@ -12029,9 +12657,9 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000001, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 1e-07, + "output_cost_per_token": 1e-07, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12042,9 +12670,9 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000001, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 1e-07, + "output_cost_per_token": 1e-07, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12055,9 +12683,9 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000001, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 1e-07, + "output_cost_per_token": 1e-07, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12068,9 +12696,9 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_vision": true, @@ -12082,9 +12710,9 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_vision": true, @@ -12095,9 +12723,9 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12108,9 +12736,9 @@ "max_tokens": 65536, "max_input_tokens": 65536, "max_output_tokens": 65536, - "input_cost_per_token": 0.0000012, - "output_cost_per_token": 0.0000012, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 1.2e-06, + "output_cost_per_token": 1.2e-06, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12121,9 +12749,9 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 32768, - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12134,9 +12762,9 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12147,9 +12775,9 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 32768, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000003, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 3e-06, + "output_cost_per_token": 3e-06, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12160,9 +12788,9 @@ "max_tokens": 65536, "max_input_tokens": 65536, "max_output_tokens": 65536, - "input_cost_per_token": 0.0000012, - "output_cost_per_token": 0.0000012, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 1.2e-06, + "output_cost_per_token": 1.2e-06, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12173,8 +12801,8 @@ "max_tokens": 8192, "max_input_tokens": 128000, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "fireworks_ai", "mode": "chat", "supports_response_schema": true, @@ -12185,8 +12813,8 @@ "max_tokens": 20480, "max_input_tokens": 128000, "max_output_tokens": 20480, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 8e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 8e-06, "litellm_provider": "fireworks_ai", "mode": "chat", "supports_response_schema": true, @@ -12197,8 +12825,8 @@ "max_tokens": 20480, "max_input_tokens": 128000, "max_output_tokens": 20480, - "input_cost_per_token": 0.55e-6, - "output_cost_per_token": 2.19e-6, + "input_cost_per_token": 5.5e-07, + "output_cost_per_token": 2.19e-06, "litellm_provider": "fireworks_ai", "mode": "chat", "supports_response_schema": true, @@ -12209,8 +12837,8 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 3e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "fireworks_ai", "mode": "chat", "supports_response_schema": true, @@ -12221,8 +12849,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.22e-6, - "output_cost_per_token": 0.88e-6, + "input_cost_per_token": 2.2e-07, + "output_cost_per_token": 8.8e-07, "litellm_provider": "fireworks_ai", "mode": "chat", "supports_response_schema": true, @@ -12233,8 +12861,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.15e-6, - "output_cost_per_token": 0.60e-6, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "fireworks_ai", "mode": "chat", "supports_response_schema": true, @@ -12244,8 +12872,8 @@ "fireworks_ai/nomic-ai/nomic-embed-text-v1.5": { "max_tokens": 8192, "max_input_tokens": 8192, - "input_cost_per_token": 0.000000008, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 8e-09, + "output_cost_per_token": 0.0, "litellm_provider": "fireworks_ai-embedding-models", "mode": "embedding", "source": "https://fireworks.ai/pricing" @@ -12253,8 +12881,8 @@ "fireworks_ai/nomic-ai/nomic-embed-text-v1": { "max_tokens": 8192, "max_input_tokens": 8192, - "input_cost_per_token": 0.000000008, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 8e-09, + "output_cost_per_token": 0.0, "litellm_provider": "fireworks_ai-embedding-models", "mode": "embedding", "source": "https://fireworks.ai/pricing" @@ -12262,8 +12890,8 @@ "fireworks_ai/WhereIsAI/UAE-Large-V1": { "max_tokens": 512, "max_input_tokens": 512, - "input_cost_per_token": 0.000000016, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.6e-08, + "output_cost_per_token": 0.0, "litellm_provider": "fireworks_ai-embedding-models", "mode": "embedding", "source": "https://fireworks.ai/pricing" @@ -12271,8 +12899,8 @@ "fireworks_ai/thenlper/gte-large": { "max_tokens": 512, "max_input_tokens": 512, - "input_cost_per_token": 0.000000016, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.6e-08, + "output_cost_per_token": 0.0, "litellm_provider": "fireworks_ai-embedding-models", "mode": "embedding", "source": "https://fireworks.ai/pricing" @@ -12280,35 +12908,35 @@ "fireworks_ai/thenlper/gte-base": { "max_tokens": 512, "max_input_tokens": 512, - "input_cost_per_token": 0.000000008, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 8e-09, + "output_cost_per_token": 0.0, "litellm_provider": "fireworks_ai-embedding-models", "mode": "embedding", "source": "https://fireworks.ai/pricing" }, "fireworks-ai-up-to-4b": { - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "fireworks_ai" }, "fireworks-ai-4.1b-to-16b": { - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "fireworks_ai" }, "fireworks-ai-above-16b": { - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "fireworks_ai" }, "fireworks-ai-moe-up-to-56b": { - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "fireworks_ai" }, "fireworks-ai-56b-to-176b": { - "input_cost_per_token": 0.0000012, - "output_cost_per_token": 0.0000012, + "input_cost_per_token": 1.2e-06, + "output_cost_per_token": 1.2e-06, "litellm_provider": "fireworks_ai" }, "fireworks-ai-default": { @@ -12317,270 +12945,270 @@ "litellm_provider": "fireworks_ai" }, "fireworks-ai-embedding-up-to-150m": { - "input_cost_per_token": 0.000000008, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 8e-09, + "output_cost_per_token": 0.0, "litellm_provider": "fireworks_ai-embedding-models" }, "fireworks-ai-embedding-150m-to-350m": { - "input_cost_per_token": 0.000000016, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.6e-08, + "output_cost_per_token": 0.0, "litellm_provider": "fireworks_ai-embedding-models" }, - "anyscale/mistralai/Mistral-7B-Instruct-v0.1": { + "anyscale/mistralai/Mistral-7B-Instruct-v0.1": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, - "litellm_provider": "anyscale", + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, + "litellm_provider": "anyscale", "mode": "chat", "supports_function_calling": true, "source": "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/mistralai-Mistral-7B-Instruct-v0.1" - }, - "anyscale/mistralai/Mixtral-8x7B-Instruct-v0.1": { + }, + "anyscale/mistralai/Mixtral-8x7B-Instruct-v0.1": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, - "litellm_provider": "anyscale", + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, + "litellm_provider": "anyscale", "mode": "chat", "supports_function_calling": true, "source": "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/mistralai-Mixtral-8x7B-Instruct-v0.1" - }, - "anyscale/mistralai/Mixtral-8x22B-Instruct-v0.1": { + }, + "anyscale/mistralai/Mixtral-8x22B-Instruct-v0.1": { "max_tokens": 65536, "max_input_tokens": 65536, "max_output_tokens": 65536, - "input_cost_per_token": 0.00000090, - "output_cost_per_token": 0.00000090, - "litellm_provider": "anyscale", + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, + "litellm_provider": "anyscale", "mode": "chat", "supports_function_calling": true, "source": "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/mistralai-Mixtral-8x22B-Instruct-v0.1" - }, - "anyscale/HuggingFaceH4/zephyr-7b-beta": { + }, + "anyscale/HuggingFaceH4/zephyr-7b-beta": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, - "litellm_provider": "anyscale", + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, + "litellm_provider": "anyscale", "mode": "chat" - }, - "anyscale/google/gemma-7b-it": { + }, + "anyscale/google/gemma-7b-it": { "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, - "litellm_provider": "anyscale", + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, + "litellm_provider": "anyscale", "mode": "chat", "source": "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/google-gemma-7b-it" - }, - "anyscale/meta-llama/Llama-2-7b-chat-hf": { + }, + "anyscale/meta-llama/Llama-2-7b-chat-hf": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, - "litellm_provider": "anyscale", + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, + "litellm_provider": "anyscale", "mode": "chat" - }, - "anyscale/meta-llama/Llama-2-13b-chat-hf": { + }, + "anyscale/meta-llama/Llama-2-13b-chat-hf": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000025, - "litellm_provider": "anyscale", + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 2.5e-07, + "litellm_provider": "anyscale", "mode": "chat" - }, - "anyscale/meta-llama/Llama-2-70b-chat-hf": { + }, + "anyscale/meta-llama/Llama-2-70b-chat-hf": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000001, - "litellm_provider": "anyscale", + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, + "litellm_provider": "anyscale", "mode": "chat" - }, - "anyscale/codellama/CodeLlama-34b-Instruct-hf": { + }, + "anyscale/codellama/CodeLlama-34b-Instruct-hf": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000001, - "litellm_provider": "anyscale", + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, + "litellm_provider": "anyscale", "mode": "chat" - }, - "anyscale/codellama/CodeLlama-70b-Instruct-hf": { + }, + "anyscale/codellama/CodeLlama-70b-Instruct-hf": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000001, - "litellm_provider": "anyscale", + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, + "litellm_provider": "anyscale", "mode": "chat", - "source" : "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/codellama-CodeLlama-70b-Instruct-hf" - }, - "anyscale/meta-llama/Meta-Llama-3-8B-Instruct": { + "source": "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/codellama-CodeLlama-70b-Instruct-hf" + }, + "anyscale/meta-llama/Meta-Llama-3-8B-Instruct": { "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, - "litellm_provider": "anyscale", + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, + "litellm_provider": "anyscale", "mode": "chat", "source": "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/meta-llama-Meta-Llama-3-8B-Instruct" - }, - "anyscale/meta-llama/Meta-Llama-3-70B-Instruct": { + }, + "anyscale/meta-llama/Meta-Llama-3-70B-Instruct": { "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000100, - "output_cost_per_token": 0.00000100, - "litellm_provider": "anyscale", + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, + "litellm_provider": "anyscale", "mode": "chat", - "source" : "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/meta-llama-Meta-Llama-3-70B-Instruct" - }, - "cloudflare/@cf/meta/llama-2-7b-chat-fp16": { - "max_tokens": 3072, - "max_input_tokens": 3072, - "max_output_tokens": 3072, - "input_cost_per_token": 0.000001923, - "output_cost_per_token": 0.000001923, - "litellm_provider": "cloudflare", + "source": "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/meta-llama-Meta-Llama-3-70B-Instruct" + }, + "cloudflare/@cf/meta/llama-2-7b-chat-fp16": { + "max_tokens": 3072, + "max_input_tokens": 3072, + "max_output_tokens": 3072, + "input_cost_per_token": 1.923e-06, + "output_cost_per_token": 1.923e-06, + "litellm_provider": "cloudflare", "mode": "chat" - }, - "cloudflare/@cf/meta/llama-2-7b-chat-int8": { - "max_tokens": 2048, - "max_input_tokens": 2048, - "max_output_tokens": 2048, - "input_cost_per_token": 0.000001923, - "output_cost_per_token": 0.000001923, - "litellm_provider": "cloudflare", + }, + "cloudflare/@cf/meta/llama-2-7b-chat-int8": { + "max_tokens": 2048, + "max_input_tokens": 2048, + "max_output_tokens": 2048, + "input_cost_per_token": 1.923e-06, + "output_cost_per_token": 1.923e-06, + "litellm_provider": "cloudflare", "mode": "chat" - }, - "cloudflare/@cf/mistral/mistral-7b-instruct-v0.1": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.000001923, - "output_cost_per_token": 0.000001923, - "litellm_provider": "cloudflare", + }, + "cloudflare/@cf/mistral/mistral-7b-instruct-v0.1": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 1.923e-06, + "output_cost_per_token": 1.923e-06, + "litellm_provider": "cloudflare", "mode": "chat" - }, - "cloudflare/@hf/thebloke/codellama-7b-instruct-awq": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000001923, - "output_cost_per_token": 0.000001923, - "litellm_provider": "cloudflare", - "mode": "chat" - }, - "voyage/voyage-01": { + }, + "cloudflare/@hf/thebloke/codellama-7b-instruct-awq": { "max_tokens": 4096, "max_input_tokens": 4096, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, + "max_output_tokens": 4096, + "input_cost_per_token": 1.923e-06, + "output_cost_per_token": 1.923e-06, + "litellm_provider": "cloudflare", + "mode": "chat" + }, + "voyage/voyage-01": { + "max_tokens": 4096, + "max_input_tokens": 4096, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-lite-01": { "max_tokens": 4096, "max_input_tokens": 4096, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-large-2": { "max_tokens": 16000, "max_input_tokens": 16000, - "input_cost_per_token": 0.00000012, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.2e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-finance-2": { "max_tokens": 32000, "max_input_tokens": 32000, - "input_cost_per_token": 0.00000012, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.2e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-lite-02-instruct": { "max_tokens": 4000, "max_input_tokens": 4000, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-law-2": { "max_tokens": 16000, "max_input_tokens": 16000, - "input_cost_per_token": 0.00000012, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.2e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-code-2": { "max_tokens": 16000, "max_input_tokens": 16000, - "input_cost_per_token": 0.00000012, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.2e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-2": { "max_tokens": 4000, "max_input_tokens": 4000, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-3-large": { "max_tokens": 32000, "max_input_tokens": 32000, - "input_cost_per_token": 0.00000018, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.8e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-3": { "max_tokens": 32000, "max_input_tokens": 32000, - "input_cost_per_token": 0.00000006, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 6e-08, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-3-lite": { "max_tokens": 32000, "max_input_tokens": 32000, - "input_cost_per_token": 0.00000002, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 2e-08, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-code-3": { "max_tokens": 32000, "max_input_tokens": 32000, - "input_cost_per_token": 0.00000018, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.8e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-multimodal-3": { "max_tokens": 32000, "max_input_tokens": 32000, - "input_cost_per_token": 0.00000012, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.2e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, @@ -12589,8 +13217,8 @@ "max_input_tokens": 16000, "max_output_tokens": 16000, "max_query_tokens": 16000, - "input_cost_per_token": 0.00000005, - "input_cost_per_query": 0.00000005, + "input_cost_per_token": 5e-08, + "input_cost_per_query": 5e-08, "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "rerank" @@ -12600,8 +13228,8 @@ "max_input_tokens": 8000, "max_output_tokens": 8000, "max_query_tokens": 8000, - "input_cost_per_token": 0.00000002, - "input_cost_per_query": 0.00000002, + "input_cost_per_token": 2e-08, + "input_cost_per_query": 2e-08, "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "rerank" @@ -12609,15 +13237,17 @@ "databricks/databricks-claude-3-7-sonnet": { "max_tokens": 200000, "max_input_tokens": 200000, - "max_output_tokens": 128000, - "input_cost_per_token": 0.0000025, - "input_dbu_cost_per_token": 0.00003571, - "output_cost_per_token": 0.000017857, + "max_output_tokens": 128000, + "input_cost_per_token": 2.5e-06, + "input_dbu_cost_per_token": 3.571e-05, + "output_cost_per_token": 1.7857e-05, "output_db_cost_per_token": 0.000214286, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Claude 3.7 conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Claude 3.7 conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_assistant_prefill": true, "supports_function_calling": true, "supports_tool_choice": true, @@ -12626,175 +13256,199 @@ "databricks/databricks-meta-llama-3-1-405b-instruct": { "max_tokens": 128000, "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 0.000005, - "input_dbu_cost_per_token": 0.000071429, - "output_cost_per_token": 0.00001500002, + "max_output_tokens": 128000, + "input_cost_per_token": 5e-06, + "input_dbu_cost_per_token": 7.1429e-05, + "output_cost_per_token": 1.500002e-05, "output_db_cost_per_token": 0.000214286, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-meta-llama-3-1-70b-instruct": { "max_tokens": 128000, "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 0.00000100002, - "input_dbu_cost_per_token": 0.000014286, - "output_cost_per_token": 0.00000299999, - "output_dbu_cost_per_token": 0.000042857, + "max_output_tokens": 128000, + "input_cost_per_token": 1.00002e-06, + "input_dbu_cost_per_token": 1.4286e-05, + "output_cost_per_token": 2.99999e-06, + "output_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-meta-llama-3-3-70b-instruct": { "max_tokens": 128000, "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 0.00000100002, - "input_dbu_cost_per_token": 0.000014286, - "output_cost_per_token": 0.00000299999, - "output_dbu_cost_per_token": 0.000042857, + "max_output_tokens": 128000, + "input_cost_per_token": 1.00002e-06, + "input_dbu_cost_per_token": 1.4286e-05, + "output_cost_per_token": 2.99999e-06, + "output_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-llama-4-maverick": { "max_tokens": 128000, "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 0.000005, - "input_dbu_cost_per_token": 0.00007143, - "output_cost_per_token": 0.000015, + "max_output_tokens": 128000, + "input_cost_per_token": 5e-06, + "input_dbu_cost_per_token": 7.143e-05, + "output_cost_per_token": 1.5e-05, "output_dbu_cost_per_token": 0.00021429, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Databricks documentation now provides both DBU costs (_dbu_cost_per_token) and dollar costs(_cost_per_token)."}, + "metadata": { + "notes": "Databricks documentation now provides both DBU costs (_dbu_cost_per_token) and dollar costs(_cost_per_token)." + }, "supports_tool_choice": true }, "databricks/databricks-dbrx-instruct": { "max_tokens": 32768, "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 0.00000074998, - "input_dbu_cost_per_token": 0.000010714, - "output_cost_per_token": 0.00000224901, - "output_dbu_cost_per_token": 0.000032143, + "max_output_tokens": 32768, + "input_cost_per_token": 7.4998e-07, + "input_dbu_cost_per_token": 1.0714e-05, + "output_cost_per_token": 2.24901e-06, + "output_dbu_cost_per_token": 3.2143e-05, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-meta-llama-3-70b-instruct": { "max_tokens": 128000, "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 0.00000100002, - "input_dbu_cost_per_token": 0.000014286, - "output_cost_per_token": 0.00000299999, - "output_dbu_cost_per_token": 0.000042857, + "max_output_tokens": 128000, + "input_cost_per_token": 1.00002e-06, + "input_dbu_cost_per_token": 1.4286e-05, + "output_cost_per_token": 2.99999e-06, + "output_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-llama-2-70b-chat": { "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000050001, - "input_dbu_cost_per_token": 0.000007143, - "output_cost_per_token": 0.0000015, - "output_dbu_cost_per_token": 0.000021429, + "max_output_tokens": 4096, + "input_cost_per_token": 5.0001e-07, + "input_dbu_cost_per_token": 7.143e-06, + "output_cost_per_token": 1.5e-06, + "output_dbu_cost_per_token": 2.1429e-05, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-mixtral-8x7b-instruct": { "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000050001, - "input_dbu_cost_per_token": 0.000007143, - "output_cost_per_token": 0.00000099902, - "output_dbu_cost_per_token": 0.000014286, + "max_output_tokens": 4096, + "input_cost_per_token": 5.0001e-07, + "input_dbu_cost_per_token": 7.143e-06, + "output_cost_per_token": 9.9902e-07, + "output_dbu_cost_per_token": 1.4286e-05, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-mpt-30b-instruct": { "max_tokens": 8192, "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000099902, - "input_dbu_cost_per_token": 0.000014286, - "output_cost_per_token": 0.00000099902, - "output_dbu_cost_per_token": 0.000014286, + "max_output_tokens": 8192, + "input_cost_per_token": 9.9902e-07, + "input_dbu_cost_per_token": 1.4286e-05, + "output_cost_per_token": 9.9902e-07, + "output_dbu_cost_per_token": 1.4286e-05, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-mpt-7b-instruct": { "max_tokens": 8192, "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000050001, - "input_dbu_cost_per_token": 0.000007143, + "max_output_tokens": 8192, + "input_cost_per_token": 5.0001e-07, + "input_dbu_cost_per_token": 7.143e-06, "output_cost_per_token": 0.0, "output_dbu_cost_per_token": 0.0, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-bge-large-en": { "max_tokens": 512, "max_input_tokens": 512, - "output_vector_size": 1024, - "input_cost_per_token": 0.00000010003, - "input_dbu_cost_per_token": 0.000001429, + "output_vector_size": 1024, + "input_cost_per_token": 1.0003e-07, + "input_dbu_cost_per_token": 1.429e-06, "output_cost_per_token": 0.0, "output_dbu_cost_per_token": 0.0, "litellm_provider": "databricks", "mode": "embedding", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."} + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + } }, "databricks/databricks-gte-large-en": { "max_tokens": 8192, "max_input_tokens": 8192, - "output_vector_size": 1024, - "input_cost_per_token": 0.00000012999, - "input_dbu_cost_per_token": 0.000001857, + "output_vector_size": 1024, + "input_cost_per_token": 1.2999e-07, + "input_dbu_cost_per_token": 1.857e-06, "output_cost_per_token": 0.0, "output_dbu_cost_per_token": 0.0, "litellm_provider": "databricks", "mode": "embedding", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."} + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + } }, "sambanova/Meta-Llama-3.1-8B-Instruct": { "max_tokens": 16384, "max_input_tokens": 16384, - "max_output_tokens": 16384, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000002, + "max_output_tokens": 16384, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "sambanova", "mode": "chat", "supports_function_calling": true, @@ -12805,9 +13459,9 @@ "sambanova/Meta-Llama-3.1-405B-Instruct": { "max_tokens": 16384, "max_input_tokens": 16384, - "max_output_tokens": 16384, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000010, + "max_output_tokens": 16384, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "sambanova", "mode": "chat", "supports_function_calling": true, @@ -12818,9 +13472,9 @@ "sambanova/Meta-Llama-3.2-1B-Instruct": { "max_tokens": 16384, "max_input_tokens": 16384, - "max_output_tokens": 16384, - "input_cost_per_token": 0.00000004, - "output_cost_per_token": 0.00000008, + "max_output_tokens": 16384, + "input_cost_per_token": 4e-08, + "output_cost_per_token": 8e-08, "litellm_provider": "sambanova", "mode": "chat", "source": "https://cloud.sambanova.ai/plans/pricing" @@ -12828,9 +13482,9 @@ "sambanova/Meta-Llama-3.2-3B-Instruct": { "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000008, - "output_cost_per_token": 0.00000016, + "max_output_tokens": 4096, + "input_cost_per_token": 8e-08, + "output_cost_per_token": 1.6e-07, "litellm_provider": "sambanova", "mode": "chat", "source": "https://cloud.sambanova.ai/plans/pricing" @@ -12838,9 +13492,9 @@ "sambanova/Llama-4-Maverick-17B-128E-Instruct": { "max_tokens": 131072, "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 0.00000063, - "output_cost_per_token": 0.0000018, + "max_output_tokens": 131072, + "input_cost_per_token": 6.3e-07, + "output_cost_per_token": 1.8e-06, "litellm_provider": "sambanova", "mode": "chat", "supports_function_calling": true, @@ -12848,28 +13502,32 @@ "supports_response_schema": true, "supports_vision": true, "source": "https://cloud.sambanova.ai/plans/pricing", - "metadata": {"notes": "For vision models, images are converted to 6432 input tokens and are billed at that amount"} + "metadata": { + "notes": "For vision models, images are converted to 6432 input tokens and are billed at that amount" + } }, "sambanova/Llama-4-Scout-17B-16E-Instruct": { "max_tokens": 8192, "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.0000004, - "output_cost_per_token": 0.0000007, + "max_output_tokens": 8192, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "sambanova", "mode": "chat", "supports_function_calling": true, "supports_tool_choice": true, "supports_response_schema": true, "source": "https://cloud.sambanova.ai/plans/pricing", - "metadata": {"notes": "For vision models, images are converted to 6432 input tokens and are billed at that amount"} + "metadata": { + "notes": "For vision models, images are converted to 6432 input tokens and are billed at that amount" + } }, "sambanova/Meta-Llama-3.3-70B-Instruct": { "max_tokens": 131072, "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.0000012, + "max_output_tokens": 131072, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 1.2e-06, "litellm_provider": "sambanova", "mode": "chat", "supports_function_calling": true, @@ -12880,9 +13538,9 @@ "sambanova/Meta-Llama-Guard-3-8B": { "max_tokens": 16384, "max_input_tokens": 16384, - "max_output_tokens": 16384, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000003, + "max_output_tokens": 16384, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 3e-07, "litellm_provider": "sambanova", "mode": "chat", "source": "https://cloud.sambanova.ai/plans/pricing" @@ -12890,9 +13548,9 @@ "sambanova/Qwen3-32B": { "max_tokens": 8192, "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.0000004, - "output_cost_per_token": 0.0000008, + "max_output_tokens": 8192, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 8e-07, "litellm_provider": "sambanova", "supports_function_calling": true, "supports_tool_choice": true, @@ -12903,9 +13561,9 @@ "sambanova/QwQ-32B": { "max_tokens": 16384, "max_input_tokens": 16384, - "max_output_tokens": 16384, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000010, + "max_output_tokens": 16384, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1e-06, "litellm_provider": "sambanova", "mode": "chat", "source": "https://cloud.sambanova.ai/plans/pricing" @@ -12913,8 +13571,8 @@ "sambanova/Qwen2-Audio-7B-Instruct": { "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, + "max_output_tokens": 4096, + "input_cost_per_token": 5e-07, "output_cost_per_token": 0.0001, "litellm_provider": "sambanova", "mode": "chat", @@ -12924,9 +13582,9 @@ "sambanova/DeepSeek-R1-Distill-Llama-70B": { "max_tokens": 131072, "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 0.0000007, - "output_cost_per_token": 0.0000014, + "max_output_tokens": 131072, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 1.4e-06, "litellm_provider": "sambanova", "mode": "chat", "source": "https://cloud.sambanova.ai/plans/pricing" @@ -12934,9 +13592,9 @@ "sambanova/DeepSeek-R1": { "max_tokens": 32768, "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000007, + "max_output_tokens": 32768, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 7e-06, "litellm_provider": "sambanova", "mode": "chat", "source": "https://cloud.sambanova.ai/plans/pricing" @@ -12944,9 +13602,9 @@ "sambanova/DeepSeek-V3-0324": { "max_tokens": 32768, "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 0.0000030, - "output_cost_per_token": 0.0000045, + "max_output_tokens": 32768, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 4.5e-06, "litellm_provider": "sambanova", "mode": "chat", "supports_function_calling": true, @@ -12957,13 +13615,13 @@ "assemblyai/nano": { "mode": "audio_transcription", "input_cost_per_second": 0.00010278, - "output_cost_per_second": 0.00, + "output_cost_per_second": 0.0, "litellm_provider": "assemblyai" }, "assemblyai/best": { "mode": "audio_transcription", - "input_cost_per_second": 0.00003333, - "output_cost_per_second": 0.00, + "input_cost_per_second": 3.333e-05, + "output_cost_per_second": 0.0, "litellm_provider": "assemblyai" }, "jina-reranker-v2-base-multilingual": { @@ -12971,8 +13629,8 @@ "max_input_tokens": 1024, "max_output_tokens": 1024, "max_document_chunks_per_query": 2048, - "input_cost_per_token": 0.000000018, - "output_cost_per_token": 0.000000018, + "input_cost_per_token": 1.8e-08, + "output_cost_per_token": 1.8e-08, "litellm_provider": "jina_ai", "mode": "rerank" }, @@ -13147,43 +13805,43 @@ "mode": "chat" }, "nscale/meta-llama/Llama-4-Scout-17B-16E-Instruct": { - "input_cost_per_token": 9e-8, - "output_cost_per_token": 2.9e-7, + "input_cost_per_token": 9e-08, + "output_cost_per_token": 2.9e-07, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" }, "nscale/Qwen/Qwen2.5-Coder-3B-Instruct": { - "input_cost_per_token": 1e-8, - "output_cost_per_token": 3e-8, + "input_cost_per_token": 1e-08, + "output_cost_per_token": 3e-08, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" }, "nscale/Qwen/Qwen2.5-Coder-7B-Instruct": { - "input_cost_per_token": 1e-8, - "output_cost_per_token": 3e-8, + "input_cost_per_token": 1e-08, + "output_cost_per_token": 3e-08, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" }, "nscale/Qwen/Qwen2.5-Coder-32B-Instruct": { - "input_cost_per_token": 6e-8, - "output_cost_per_token": 2e-7, + "input_cost_per_token": 6e-08, + "output_cost_per_token": 2e-07, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" }, "nscale/Qwen/QwQ-32B": { - "input_cost_per_token": 1.8e-7, - "output_cost_per_token": 2e-7, + "input_cost_per_token": 1.8e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" }, "nscale/deepseek-ai/DeepSeek-R1-Distill-Llama-70B": { - "input_cost_per_token": 3.75e-7, - "output_cost_per_token": 3.75e-7, + "input_cost_per_token": 3.75e-07, + "output_cost_per_token": 3.75e-07, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13192,8 +13850,8 @@ } }, "nscale/deepseek-ai/DeepSeek-R1-Distill-Llama-8B": { - "input_cost_per_token": 2.5e-8, - "output_cost_per_token": 2.5e-8, + "input_cost_per_token": 2.5e-08, + "output_cost_per_token": 2.5e-08, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13202,8 +13860,8 @@ } }, "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": { - "input_cost_per_token": 9e-8, - "output_cost_per_token": 9e-8, + "input_cost_per_token": 9e-08, + "output_cost_per_token": 9e-08, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13212,8 +13870,8 @@ } }, "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": { - "input_cost_per_token": 2e-7, - "output_cost_per_token": 2e-7, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13222,8 +13880,8 @@ } }, "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": { - "input_cost_per_token": 7e-8, - "output_cost_per_token": 7e-8, + "input_cost_per_token": 7e-08, + "output_cost_per_token": 7e-08, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13232,8 +13890,8 @@ } }, "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": { - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 1.5e-7, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13242,8 +13900,8 @@ } }, "nscale/mistralai/mixtral-8x22b-instruct-v0.1": { - "input_cost_per_token": 6e-7, - "output_cost_per_token": 6e-7, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13252,8 +13910,8 @@ } }, "nscale/meta-llama/Llama-3.1-8B-Instruct": { - "input_cost_per_token": 3e-8, - "output_cost_per_token": 3e-8, + "input_cost_per_token": 3e-08, + "output_cost_per_token": 3e-08, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13262,8 +13920,8 @@ } }, "nscale/meta-llama/Llama-3.3-70B-Instruct": { - "input_cost_per_token": 2e-7, - "output_cost_per_token": 2e-7, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13273,7 +13931,7 @@ }, "nscale/black-forest-labs/FLUX.1-schnell": { "mode": "image_generation", - "input_cost_per_pixel": 1.3e-9, + "input_cost_per_pixel": 1.3e-09, "output_cost_per_pixel": 0.0, "litellm_provider": "nscale", "supported_endpoints": [ @@ -13283,7 +13941,7 @@ }, "nscale/stabilityai/stable-diffusion-xl-base-1.0": { "mode": "image_generation", - "input_cost_per_pixel": 3e-9, + "input_cost_per_pixel": 3e-09, "output_cost_per_pixel": 0.0, "litellm_provider": "nscale", "supported_endpoints": [ @@ -13305,4 +13963,4 @@ "litellm_provider": "featherless_ai", "mode": "chat" } -} +} \ No newline at end of file diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 47c6c0f754f..988000c6401 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -1,17 +1,17 @@ { "sample_spec": { - "max_tokens": "LEGACY parameter. set to max_output_tokens if provider specifies it. IF not set to max_input_tokens, if provider specifies it.", + "max_tokens": "LEGACY parameter. set to max_output_tokens if provider specifies it. IF not set to max_input_tokens, if provider specifies it.", "max_input_tokens": "max input tokens, if the provider specifies it. if not default to max_tokens", - "max_output_tokens": "max output tokens, if the provider specifies it. if not default to max_tokens", - "input_cost_per_token": 0.0000, - "output_cost_per_token": 0.000, - "output_cost_per_reasoning_token": 0.000, + "max_output_tokens": "max output tokens, if the provider specifies it. if not default to max_tokens", + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, + "output_cost_per_reasoning_token": 0.0, "litellm_provider": "one of https://docs.litellm.ai/docs/providers", "mode": "one of: chat, embedding, completion, image_generation, audio_transcription, audio_speech, image_generation, moderation, rerank", "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_vision": true, - "supports_audio_input": true, + "supports_audio_input": true, "supports_audio_output": true, "supports_prompt_caching": true, "supports_response_schema": true, @@ -19,16 +19,16 @@ "supports_reasoning": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 0.0000, - "search_context_size_medium": 0.0000, - "search_context_size_high": 0.0000 + "search_context_size_low": 0.0, + "search_context_size_medium": 0.0, + "search_context_size_high": 0.0 }, "deprecation_date": "date when the model becomes deprecated in the format YYYY-MM-DD" }, "omni-moderation-latest": { "max_tokens": 32768, "max_input_tokens": 32768, - "max_output_tokens": 0, + "max_output_tokens": 0, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "openai", @@ -37,7 +37,7 @@ "omni-moderation-latest-intents": { "max_tokens": 32768, "max_input_tokens": 32768, - "max_output_tokens": 0, + "max_output_tokens": 0, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "openai", @@ -46,18 +46,18 @@ "omni-moderation-2024-09-26": { "max_tokens": 32768, "max_input_tokens": 32768, - "max_output_tokens": 0, + "max_output_tokens": 0, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "openai", "mode": "moderation" }, "gpt-4": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 8192, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.00006, + "max_output_tokens": 4096, + "input_cost_per_token": 3e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -69,16 +69,25 @@ "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 2e-6, - "output_cost_per_token": 8e-6, - "input_cost_per_token_batches": 1e-6, - "output_cost_per_token_batches": 4e-6, - "cache_read_input_token_cost": 0.5e-6, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, + "input_cost_per_token_batches": 1e-06, + "output_cost_per_token_batches": 4e-06, + "cache_read_input_token_cost": 5e-07, "litellm_provider": "openai", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_pdf_input": true, "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -90,25 +99,34 @@ "supports_native_streaming": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 30e-3, - "search_context_size_medium": 35e-3, - "search_context_size_high": 50e-3 + "search_context_size_low": 0.03, + "search_context_size_medium": 0.035, + "search_context_size_high": 0.05 } }, "gpt-4.1-2025-04-14": { "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 2e-6, - "output_cost_per_token": 8e-6, - "input_cost_per_token_batches": 1e-6, - "output_cost_per_token_batches": 4e-6, - "cache_read_input_token_cost": 0.5e-6, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, + "input_cost_per_token_batches": 1e-06, + "output_cost_per_token_batches": 4e-06, + "cache_read_input_token_cost": 5e-07, "litellm_provider": "openai", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_pdf_input": true, "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -120,25 +138,34 @@ "supports_native_streaming": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 30e-3, - "search_context_size_medium": 35e-3, - "search_context_size_high": 50e-3 + "search_context_size_low": 0.03, + "search_context_size_medium": 0.035, + "search_context_size_high": 0.05 } }, "gpt-4.1-mini": { "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 0.4e-6, - "output_cost_per_token": 1.6e-6, - "input_cost_per_token_batches": 0.2e-6, - "output_cost_per_token_batches": 0.8e-6, - "cache_read_input_token_cost": 0.1e-6, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 1.6e-06, + "input_cost_per_token_batches": 2e-07, + "output_cost_per_token_batches": 8e-07, + "cache_read_input_token_cost": 1e-07, "litellm_provider": "openai", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_pdf_input": true, "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -150,25 +177,34 @@ "supports_native_streaming": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 25e-3, - "search_context_size_medium": 27.5e-3, - "search_context_size_high": 30e-3 + "search_context_size_low": 0.025, + "search_context_size_medium": 0.0275, + "search_context_size_high": 0.03 } }, "gpt-4.1-mini-2025-04-14": { "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 0.4e-6, - "output_cost_per_token": 1.6e-6, - "input_cost_per_token_batches": 0.2e-6, - "output_cost_per_token_batches": 0.8e-6, - "cache_read_input_token_cost": 0.1e-6, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 1.6e-06, + "input_cost_per_token_batches": 2e-07, + "output_cost_per_token_batches": 8e-07, + "cache_read_input_token_cost": 1e-07, "litellm_provider": "openai", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_pdf_input": true, "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -180,25 +216,34 @@ "supports_native_streaming": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 25e-3, - "search_context_size_medium": 27.5e-3, - "search_context_size_high": 30e-3 + "search_context_size_low": 0.025, + "search_context_size_medium": 0.0275, + "search_context_size_high": 0.03 } }, "gpt-4.1-nano": { "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 0.1e-6, - "output_cost_per_token": 0.4e-6, - "input_cost_per_token_batches": 0.05e-6, - "output_cost_per_token_batches": 0.2e-6, - "cache_read_input_token_cost": 0.025e-6, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, + "input_cost_per_token_batches": 5e-08, + "output_cost_per_token_batches": 2e-07, + "cache_read_input_token_cost": 2.5e-08, "litellm_provider": "openai", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_pdf_input": true, "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -213,16 +258,25 @@ "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 0.1e-6, - "output_cost_per_token": 0.4e-6, - "input_cost_per_token_batches": 0.05e-6, - "output_cost_per_token_batches": 0.2e-6, - "cache_read_input_token_cost": 0.025e-6, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, + "input_cost_per_token_batches": 5e-08, + "output_cost_per_token_batches": 2e-07, + "cache_read_input_token_cost": 2.5e-08, "litellm_provider": "openai", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_pdf_input": true, "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -237,11 +291,11 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000010, - "input_cost_per_token_batches": 0.00000125, - "output_cost_per_token_batches": 0.00000500, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "input_cost_per_token_batches": 1.25e-06, + "output_cost_per_token_batches": 5e-06, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -254,38 +308,38 @@ "supports_tool_choice": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 0.030, + "search_context_size_low": 0.03, "search_context_size_medium": 0.035, - "search_context_size_high": 0.050 + "search_context_size_high": 0.05 } }, "watsonx/ibm/granite-3-8b-instruct": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 1024, - "input_cost_per_token": 0.0002, - "output_cost_per_token": 0.0002, - "litellm_provider": "watsonx", - "mode": "chat", - "supports_function_calling": true, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 1024, + "input_cost_per_token": 0.0002, + "output_cost_per_token": 0.0002, + "litellm_provider": "watsonx", + "mode": "chat", + "supports_function_calling": true, "supports_tool_choice": true, - "supports_parallel_function_calling": false, - "supports_vision": false, - "supports_audio_input": false, - "supports_audio_output": false, - "supports_prompt_caching": true, - "supports_response_schema": true, + "supports_parallel_function_calling": false, + "supports_vision": false, + "supports_audio_input": false, + "supports_audio_output": false, + "supports_prompt_caching": true, + "supports_response_schema": true, "supports_system_messages": true }, "gpt-4o-search-preview-2025-03-11": { "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000010, - "input_cost_per_token_batches": 0.00000125, - "output_cost_per_token_batches": 0.00000500, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "input_cost_per_token_batches": 1.25e-06, + "output_cost_per_token_batches": 5e-06, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -298,20 +352,20 @@ "supports_tool_choice": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 0.030, + "search_context_size_low": 0.03, "search_context_size_medium": 0.035, - "search_context_size_high": 0.050 + "search_context_size_high": 0.05 } - }, + }, "gpt-4o-search-preview": { - "max_tokens": 16384, + "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000010, - "input_cost_per_token_batches": 0.00000125, - "output_cost_per_token_batches": 0.00000500, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "input_cost_per_token_batches": 1.25e-06, + "output_cost_per_token_batches": 5e-06, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -324,20 +378,20 @@ "supports_tool_choice": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 0.030, + "search_context_size_low": 0.03, "search_context_size_medium": 0.035, - "search_context_size_high": 0.050 + "search_context_size_high": 0.05 } }, "gpt-4.5-preview": { "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.000075, + "input_cost_per_token": 7.5e-05, "output_cost_per_token": 0.00015, - "input_cost_per_token_batches": 0.0000375, - "output_cost_per_token_batches": 0.000075, - "cache_read_input_token_cost": 0.0000375, + "input_cost_per_token_batches": 3.75e-05, + "output_cost_per_token_batches": 7.5e-05, + "cache_read_input_token_cost": 3.75e-05, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -353,11 +407,11 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.000075, + "input_cost_per_token": 7.5e-05, "output_cost_per_token": 0.00015, - "input_cost_per_token_batches": 0.0000375, - "output_cost_per_token_batches": 0.000075, - "cache_read_input_token_cost": 0.0000375, + "input_cost_per_token_batches": 3.75e-05, + "output_cost_per_token_batches": 7.5e-05, + "cache_read_input_token_cost": 3.75e-05, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -374,9 +428,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, + "input_cost_per_token": 2.5e-06, "input_cost_per_audio_token": 0.0001, - "output_cost_per_token": 0.000010, + "output_cost_per_token": 1e-05, "output_cost_per_audio_token": 0.0002, "litellm_provider": "openai", "mode": "chat", @@ -391,10 +445,10 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "input_cost_per_audio_token": 0.00004, - "output_cost_per_token": 0.000010, - "output_cost_per_audio_token": 0.00008, + "input_cost_per_token": 2.5e-06, + "input_cost_per_audio_token": 4e-05, + "output_cost_per_token": 1e-05, + "output_cost_per_audio_token": 8e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -408,9 +462,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, + "input_cost_per_token": 2.5e-06, "input_cost_per_audio_token": 0.0001, - "output_cost_per_token": 0.000010, + "output_cost_per_token": 1e-05, "output_cost_per_audio_token": 0.0002, "litellm_provider": "openai", "mode": "chat", @@ -425,10 +479,10 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "input_cost_per_audio_token": 0.00001, - "output_cost_per_token": 0.0000006, - "output_cost_per_audio_token": 0.00002, + "input_cost_per_token": 1.5e-07, + "input_cost_per_audio_token": 1e-05, + "output_cost_per_token": 6e-07, + "output_cost_per_audio_token": 2e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -442,10 +496,10 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "input_cost_per_audio_token": 0.00001, - "output_cost_per_token": 0.0000006, - "output_cost_per_audio_token": 0.00002, + "input_cost_per_token": 1.5e-07, + "input_cost_per_audio_token": 1e-05, + "output_cost_per_token": 6e-07, + "output_cost_per_audio_token": 2e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -459,37 +513,11 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000060, - "input_cost_per_token_batches": 0.000000075, - "output_cost_per_token_batches": 0.00000030, - "cache_read_input_token_cost": 0.000000075, - "litellm_provider": "openai", - "mode": "chat", - "supports_pdf_input": true, - "supports_function_calling": true, - "supports_parallel_function_calling": true, - "supports_response_schema": true, - "supports_vision": true, - "supports_prompt_caching": true, - "supports_system_messages": true, - "supports_tool_choice": true, - "supports_web_search": true, - "search_context_cost_per_query": { - "search_context_size_low": 0.025, - "search_context_size_medium": 0.0275, - "search_context_size_high": 0.030 - } - }, - "gpt-4o-mini-search-preview-2025-03-11":{ - "max_tokens": 16384, - "max_input_tokens": 128000, - "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000060, - "input_cost_per_token_batches": 0.000000075, - "output_cost_per_token_batches": 0.00000030, - "cache_read_input_token_cost": 0.000000075, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "input_cost_per_token_batches": 7.5e-08, + "output_cost_per_token_batches": 3e-07, + "cache_read_input_token_cost": 7.5e-08, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -504,18 +532,44 @@ "search_context_cost_per_query": { "search_context_size_low": 0.025, "search_context_size_medium": 0.0275, - "search_context_size_high": 0.030 + "search_context_size_high": 0.03 + } + }, + "gpt-4o-mini-search-preview-2025-03-11": { + "max_tokens": 16384, + "max_input_tokens": 128000, + "max_output_tokens": 16384, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "input_cost_per_token_batches": 7.5e-08, + "output_cost_per_token_batches": 3e-07, + "cache_read_input_token_cost": 7.5e-08, + "litellm_provider": "openai", + "mode": "chat", + "supports_pdf_input": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_response_schema": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_web_search": true, + "search_context_cost_per_query": { + "search_context_size_low": 0.025, + "search_context_size_medium": 0.0275, + "search_context_size_high": 0.03 } }, "gpt-4o-mini-search-preview": { "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000060, - "input_cost_per_token_batches": 0.000000075, - "output_cost_per_token_batches": 0.00000030, - "cache_read_input_token_cost": 0.000000075, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "input_cost_per_token_batches": 7.5e-08, + "output_cost_per_token_batches": 3e-07, + "cache_read_input_token_cost": 7.5e-08, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -530,18 +584,18 @@ "search_context_cost_per_query": { "search_context_size_low": 0.025, "search_context_size_medium": 0.0275, - "search_context_size_high": 0.030 + "search_context_size_high": 0.03 } }, "gpt-4o-mini-2024-07-18": { "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000060, - "input_cost_per_token_batches": 0.000000075, - "output_cost_per_token_batches": 0.00000030, - "cache_read_input_token_cost": 0.000000075, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "input_cost_per_token_batches": 7.5e-08, + "output_cost_per_token_batches": 3e-07, + "cache_read_input_token_cost": 7.5e-08, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -553,9 +607,9 @@ "supports_system_messages": true, "supports_tool_choice": true, "search_context_cost_per_query": { - "search_context_size_low": 30.00, - "search_context_size_medium": 35.00, - "search_context_size_high": 50.00 + "search_context_size_low": 30.0, + "search_context_size_medium": 35.0, + "search_context_size_high": 50.0 } }, "o1-pro": { @@ -564,7 +618,7 @@ "max_output_tokens": 100000, "input_cost_per_token": 0.00015, "output_cost_per_token": 0.0006, - "input_cost_per_token_batches": 0.000075, + "input_cost_per_token_batches": 7.5e-05, "output_cost_per_token_batches": 0.0003, "litellm_provider": "openai", "mode": "responses", @@ -578,9 +632,17 @@ "supports_tool_choice": true, "supports_native_streaming": false, "supports_reasoning": true, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], - "supported_endpoints": ["/v1/responses", "/v1/batch"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supported_endpoints": [ + "/v1/responses", + "/v1/batch" + ] }, "o1-pro-2025-03-19": { "max_tokens": 100000, @@ -588,7 +650,7 @@ "max_output_tokens": 100000, "input_cost_per_token": 0.00015, "output_cost_per_token": 0.0006, - "input_cost_per_token_batches": 0.000075, + "input_cost_per_token_batches": 7.5e-05, "output_cost_per_token_batches": 0.0003, "litellm_provider": "openai", "mode": "responses", @@ -602,17 +664,25 @@ "supports_tool_choice": true, "supports_native_streaming": false, "supports_reasoning": true, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], - "supported_endpoints": ["/v1/responses", "/v1/batch"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supported_endpoints": [ + "/v1/responses", + "/v1/batch" + ] }, "o1": { "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.00006, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -629,9 +699,9 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.0000044, - "cache_read_input_token_cost": 0.00000055, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 5.5e-07, "litellm_provider": "openai", "mode": "chat", "supports_vision": true, @@ -642,13 +712,20 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 12e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.2e-05, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -662,9 +739,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 1e-5, - "output_cost_per_token": 4e-5, - "cache_read_input_token_cost": 2.5e-6, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 4e-05, + "cache_read_input_token_cost": 2.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -680,9 +757,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 1e-5, - "output_cost_per_token": 4e-5, - "cache_read_input_token_cost": 2.5e-6, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 4e-05, + "cache_read_input_token_cost": 2.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -698,9 +775,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.0000044, - "cache_read_input_token_cost": 0.00000055, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 5.5e-07, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -715,9 +792,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.0000044, - "cache_read_input_token_cost": 0.00000055, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 5.5e-07, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -732,9 +809,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 1.1e-6, - "output_cost_per_token": 4.4e-6, - "cache_read_input_token_cost": 2.75e-7, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 2.75e-07, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -750,9 +827,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 1.1e-6, - "output_cost_per_token": 4.4e-6, - "cache_read_input_token_cost": 2.75e-7, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 2.75e-07, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -768,9 +845,9 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000012, - "cache_read_input_token_cost": 0.0000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.2e-05, + "cache_read_input_token_cost": 1.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -782,9 +859,9 @@ "max_tokens": 32768, "max_input_tokens": 128000, "max_output_tokens": 32768, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -796,9 +873,9 @@ "max_tokens": 32768, "max_input_tokens": 128000, "max_output_tokens": 32768, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -810,9 +887,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -829,8 +906,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -845,10 +922,10 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000015, - "input_cost_per_token_batches": 0.0000025, - "output_cost_per_token_batches": 0.0000075, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 1.5e-05, + "input_cost_per_token_batches": 2.5e-06, + "output_cost_per_token_batches": 7.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -863,11 +940,11 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000010, - "input_cost_per_token_batches": 0.00000125, - "output_cost_per_token_batches": 0.0000050, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "input_cost_per_token_batches": 1.25e-06, + "output_cost_per_token_batches": 5e-06, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -880,20 +957,20 @@ "supports_tool_choice": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 0.030, + "search_context_size_low": 0.03, "search_context_size_medium": 0.035, - "search_context_size_high": 0.050 + "search_context_size_high": 0.05 } }, "gpt-4o-2024-11-20": { "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000010, - "input_cost_per_token_batches": 0.00000125, - "output_cost_per_token_batches": 0.0000050, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "input_cost_per_token_batches": 1.25e-06, + "output_cost_per_token_batches": 5e-06, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -909,11 +986,11 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, + "input_cost_per_token": 5e-06, "input_cost_per_audio_token": 0.0001, - "cache_read_input_token_cost": 0.0000025, - "cache_creation_input_audio_token_cost": 0.00002, - "output_cost_per_token": 0.00002, + "cache_read_input_token_cost": 2.5e-06, + "cache_creation_input_audio_token_cost": 2e-05, + "output_cost_per_token": 2e-05, "output_cost_per_audio_token": 0.0002, "litellm_provider": "openai", "mode": "chat", @@ -928,11 +1005,11 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, - "input_cost_per_audio_token": 0.00004, - "cache_read_input_token_cost": 0.0000025, - "output_cost_per_token": 0.00002, - "output_cost_per_audio_token": 0.00008, + "input_cost_per_token": 5e-06, + "input_cost_per_audio_token": 4e-05, + "cache_read_input_token_cost": 2.5e-06, + "output_cost_per_token": 2e-05, + "output_cost_per_audio_token": 8e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -946,11 +1023,11 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, - "input_cost_per_audio_token": 0.00004, - "cache_read_input_token_cost": 0.0000025, - "output_cost_per_token": 0.00002, - "output_cost_per_audio_token": 0.00008, + "input_cost_per_token": 5e-06, + "input_cost_per_audio_token": 4e-05, + "cache_read_input_token_cost": 2.5e-06, + "output_cost_per_token": 2e-05, + "output_cost_per_audio_token": 8e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -964,12 +1041,12 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000006, - "input_cost_per_audio_token": 0.00001, - "cache_read_input_token_cost": 0.0000003, - "cache_creation_input_audio_token_cost": 0.0000003, - "output_cost_per_token": 0.0000024, - "output_cost_per_audio_token": 0.00002, + "input_cost_per_token": 6e-07, + "input_cost_per_audio_token": 1e-05, + "cache_read_input_token_cost": 3e-07, + "cache_creation_input_audio_token_cost": 3e-07, + "output_cost_per_token": 2.4e-06, + "output_cost_per_audio_token": 2e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -983,12 +1060,12 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000006, - "input_cost_per_audio_token": 0.00001, - "cache_read_input_token_cost": 0.0000003, - "cache_creation_input_audio_token_cost": 0.0000003, - "output_cost_per_token": 0.0000024, - "output_cost_per_audio_token": 0.00002, + "input_cost_per_token": 6e-07, + "input_cost_per_audio_token": 1e-05, + "cache_read_input_token_cost": 3e-07, + "cache_creation_input_audio_token_cost": 3e-07, + "output_cost_per_token": 2.4e-06, + "output_cost_per_audio_token": 2e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1002,8 +1079,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -1017,8 +1094,8 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.00006, + "input_cost_per_token": 3e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "openai", "mode": "chat", "supports_prompt_caching": true, @@ -1029,8 +1106,8 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.00006, + "input_cost_per_token": 3e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1043,7 +1120,7 @@ "max_tokens": 4096, "max_input_tokens": 32768, "max_output_tokens": 4096, - "input_cost_per_token": 0.00006, + "input_cost_per_token": 6e-05, "output_cost_per_token": 0.00012, "litellm_provider": "openai", "mode": "chat", @@ -1055,7 +1132,7 @@ "max_tokens": 4096, "max_input_tokens": 32768, "max_output_tokens": 4096, - "input_cost_per_token": 0.00006, + "input_cost_per_token": 6e-05, "output_cost_per_token": 0.00012, "litellm_provider": "openai", "mode": "chat", @@ -1067,7 +1144,7 @@ "max_tokens": 4096, "max_input_tokens": 32768, "max_output_tokens": 4096, - "input_cost_per_token": 0.00006, + "input_cost_per_token": 6e-05, "output_cost_per_token": 0.00012, "litellm_provider": "openai", "mode": "chat", @@ -1079,8 +1156,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -1095,8 +1172,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -1111,8 +1188,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1125,8 +1202,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1139,8 +1216,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "openai", "mode": "chat", "supports_vision": true, @@ -1154,8 +1231,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "openai", "mode": "chat", "supports_vision": true, @@ -1169,8 +1246,8 @@ "max_tokens": 4097, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1182,8 +1259,8 @@ "max_tokens": 4097, "max_input_tokens": 4097, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "openai", "mode": "chat", "supports_prompt_caching": true, @@ -1194,8 +1271,8 @@ "max_tokens": 4097, "max_input_tokens": 4097, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1207,8 +1284,8 @@ "max_tokens": 16385, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000010, - "output_cost_per_token": 0.0000020, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1221,8 +1298,8 @@ "max_tokens": 16385, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1235,8 +1312,8 @@ "max_tokens": 16385, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000004, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 4e-06, "litellm_provider": "openai", "mode": "chat", "supports_prompt_caching": true, @@ -1247,8 +1324,8 @@ "max_tokens": 16385, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000004, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 4e-06, "litellm_provider": "openai", "mode": "chat", "supports_prompt_caching": true, @@ -1259,10 +1336,10 @@ "max_tokens": 4096, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000006, - "input_cost_per_token_batches": 0.0000015, - "output_cost_per_token_batches": 0.000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 6e-06, + "input_cost_per_token_batches": 1.5e-06, + "output_cost_per_token_batches": 3e-06, "litellm_provider": "openai", "mode": "chat", "supports_system_messages": true, @@ -1272,8 +1349,8 @@ "max_tokens": 4096, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "openai", "mode": "chat", "supports_system_messages": true, @@ -1283,8 +1360,8 @@ "max_tokens": 4096, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "openai", "mode": "chat", "supports_system_messages": true, @@ -1294,8 +1371,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "openai", "mode": "chat", "supports_system_messages": true, @@ -1305,8 +1382,8 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.00006, + "input_cost_per_token": 3e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1318,10 +1395,10 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000375, - "output_cost_per_token": 0.000015, - "input_cost_per_token_batches": 0.000001875, - "output_cost_per_token_batches": 0.000007500, + "input_cost_per_token": 3.75e-06, + "output_cost_per_token": 1.5e-05, + "input_cost_per_token_batches": 1.875e-06, + "output_cost_per_token_batches": 7.5e-06, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -1336,9 +1413,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000375, - "cache_creation_input_token_cost": 0.000001875, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3.75e-06, + "cache_creation_input_token_cost": 1.875e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "openai", "mode": "chat", "supports_pdf_input": true, @@ -1354,11 +1431,11 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000012, - "input_cost_per_token_batches": 0.000000150, - "output_cost_per_token_batches": 0.000000600, - "cache_read_input_token_cost": 0.00000015, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 1.2e-06, + "input_cost_per_token_batches": 1.5e-07, + "output_cost_per_token_batches": 6e-07, + "cache_read_input_token_cost": 1.5e-07, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, @@ -1374,10 +1451,10 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 4096, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000002, - "input_cost_per_token_batches": 0.000001, - "output_cost_per_token_batches": 0.000001, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 2e-06, + "input_cost_per_token_batches": 1e-06, + "output_cost_per_token_batches": 1e-06, "litellm_provider": "text-completion-openai", "mode": "completion" }, @@ -1385,10 +1462,10 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000004, - "output_cost_per_token": 0.0000004, - "input_cost_per_token_batches": 0.0000002, - "output_cost_per_token_batches": 0.0000002, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 4e-07, + "input_cost_per_token_batches": 2e-07, + "output_cost_per_token_batches": 2e-07, "litellm_provider": "text-completion-openai", "mode": "completion" }, @@ -1396,40 +1473,40 @@ "max_tokens": 8191, "max_input_tokens": 8191, "output_vector_size": 3072, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.000000, - "input_cost_per_token_batches": 0.000000065, - "output_cost_per_token_batches": 0.000000000, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 0.0, + "input_cost_per_token_batches": 6.5e-08, + "output_cost_per_token_batches": 0.0, "litellm_provider": "openai", "mode": "embedding" }, "text-embedding-3-small": { "max_tokens": 8191, "max_input_tokens": 8191, - "output_vector_size": 1536, - "input_cost_per_token": 0.00000002, - "output_cost_per_token": 0.000000, - "input_cost_per_token_batches": 0.000000010, - "output_cost_per_token_batches": 0.000000000, + "output_vector_size": 1536, + "input_cost_per_token": 2e-08, + "output_cost_per_token": 0.0, + "input_cost_per_token_batches": 1e-08, + "output_cost_per_token_batches": 0.0, "litellm_provider": "openai", "mode": "embedding" }, "text-embedding-ada-002": { "max_tokens": 8191, "max_input_tokens": 8191, - "output_vector_size": 1536, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, + "output_vector_size": 1536, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "openai", "mode": "embedding" }, "text-embedding-ada-002-v2": { "max_tokens": 8191, "max_input_tokens": 8191, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, - "input_cost_per_token_batches": 0.000000050, - "output_cost_per_token_batches": 0.000000000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, + "input_cost_per_token_batches": 5e-08, + "output_cost_per_token_batches": 0.0, "litellm_provider": "openai", "mode": "embedding" }, @@ -1437,8 +1514,8 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 0, - "input_cost_per_token": 0.000000, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "openai", "mode": "moderation" }, @@ -1446,8 +1523,8 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 0, - "input_cost_per_token": 0.000000, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "openai", "mode": "moderation" }, @@ -1455,207 +1532,258 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 0, - "input_cost_per_token": 0.000000, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "openai", "mode": "moderation" }, "256-x-256/dall-e-2": { "mode": "image_generation", - "input_cost_per_pixel": 0.00000024414, + "input_cost_per_pixel": 2.4414e-07, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "512-x-512/dall-e-2": { "mode": "image_generation", - "input_cost_per_pixel": 0.0000000686, + "input_cost_per_pixel": 6.86e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "1024-x-1024/dall-e-2": { "mode": "image_generation", - "input_cost_per_pixel": 0.000000019, + "input_cost_per_pixel": 1.9e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "hd/1024-x-1792/dall-e-3": { "mode": "image_generation", - "input_cost_per_pixel": 0.00000006539, + "input_cost_per_pixel": 6.539e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "hd/1792-x-1024/dall-e-3": { "mode": "image_generation", - "input_cost_per_pixel": 0.00000006539, + "input_cost_per_pixel": 6.539e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "hd/1024-x-1024/dall-e-3": { "mode": "image_generation", - "input_cost_per_pixel": 0.00000007629, + "input_cost_per_pixel": 7.629e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "standard/1024-x-1792/dall-e-3": { "mode": "image_generation", - "input_cost_per_pixel": 0.00000004359, + "input_cost_per_pixel": 4.359e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "standard/1792-x-1024/dall-e-3": { "mode": "image_generation", - "input_cost_per_pixel": 0.00000004359, + "input_cost_per_pixel": 4.359e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "standard/1024-x-1024/dall-e-3": { "mode": "image_generation", - "input_cost_per_pixel": 0.0000000381469, + "input_cost_per_pixel": 3.81469e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai" }, "gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 4.0054321e-8, + "input_cost_per_pixel": 4.0054321e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "low/1024-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.0490417e-8, + "input_cost_per_pixel": 1.0490417e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "medium/1024-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 4.0054321e-8, + "input_cost_per_pixel": 4.0054321e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "high/1024-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.59263611e-7, + "input_cost_per_pixel": 1.59263611e-07, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "low/1024-x-1536/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.0172526e-8, + "input_cost_per_pixel": 1.0172526e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "medium/1024-x-1536/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 4.0054321e-8, + "input_cost_per_pixel": 4.0054321e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "high/1024-x-1536/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.58945719e-7, + "input_cost_per_pixel": 1.58945719e-07, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "low/1536-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.0172526e-8, + "input_cost_per_pixel": 1.0172526e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "medium/1536-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 4.0054321e-8, + "input_cost_per_pixel": 4.0054321e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "high/1536-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.58945719e-7, + "input_cost_per_pixel": 1.58945719e-07, "output_cost_per_pixel": 0.0, "litellm_provider": "openai", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "gpt-4o-transcribe": { "mode": "audio_transcription", "max_input_tokens": 16000, "max_output_tokens": 2000, - "input_cost_per_token": 0.0000025, - "input_cost_per_audio_token": 0.000006, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2.5e-06, + "input_cost_per_audio_token": 6e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "openai", - "supported_endpoints": ["/v1/audio/transcriptions"] - }, + "supported_endpoints": [ + "/v1/audio/transcriptions" + ] + }, "gpt-4o-mini-transcribe": { "mode": "audio_transcription", "max_input_tokens": 16000, "max_output_tokens": 2000, - "input_cost_per_token": 0.00000125, - "input_cost_per_audio_token": 0.000003, - "output_cost_per_token": 0.000005, + "input_cost_per_token": 1.25e-06, + "input_cost_per_audio_token": 3e-06, + "output_cost_per_token": 5e-06, "litellm_provider": "openai", - "supported_endpoints": ["/v1/audio/transcriptions"] - }, + "supported_endpoints": [ + "/v1/audio/transcriptions" + ] + }, "whisper-1": { "mode": "audio_transcription", "input_cost_per_second": 0.0001, - "output_cost_per_second": 0.0001, + "output_cost_per_second": 0.0001, "litellm_provider": "openai", - "supported_endpoints": ["/v1/audio/transcriptions"] - }, + "supported_endpoints": [ + "/v1/audio/transcriptions" + ] + }, "tts-1": { - "mode": "audio_speech", - "input_cost_per_character": 0.000015, + "mode": "audio_speech", + "input_cost_per_character": 1.5e-05, "litellm_provider": "openai", - "supported_endpoints": ["/v1/audio/speech"] + "supported_endpoints": [ + "/v1/audio/speech" + ] }, "tts-1-hd": { - "mode": "audio_speech", - "input_cost_per_character": 0.000030, + "mode": "audio_speech", + "input_cost_per_character": 3e-05, "litellm_provider": "openai", - "supported_endpoints": ["/v1/audio/speech"] + "supported_endpoints": [ + "/v1/audio/speech" + ] }, "gpt-4o-mini-tts": { - "mode": "audio_speech", - "input_cost_per_token": 2.5e-6, - "output_cost_per_token": 10e-6, - "output_cost_per_audio_token": 12e-6, + "mode": "audio_speech", + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_audio_token": 1.2e-05, "output_cost_per_second": 0.00025, "litellm_provider": "openai", - "supported_modalities": ["text", "audio"], - "supported_output_modalities": ["audio"], - "supported_endpoints": ["/v1/audio/speech"] + "supported_modalities": [ + "text", + "audio" + ], + "supported_output_modalities": [ + "audio" + ], + "supported_endpoints": [ + "/v1/audio/speech" + ] }, "azure/gpt-4o-mini-tts": { - "mode": "audio_speech", - "input_cost_per_token": 2.5e-6, - "output_cost_per_token": 10e-6, - "output_cost_per_audio_token": 12e-6, + "mode": "audio_speech", + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_audio_token": 1.2e-05, "output_cost_per_second": 0.00025, "litellm_provider": "azure", - "supported_modalities": ["text", "audio"], - "supported_output_modalities": ["audio"], - "supported_endpoints": ["/v1/audio/speech"] + "supported_modalities": [ + "text", + "audio" + ], + "supported_output_modalities": [ + "audio" + ], + "supported_endpoints": [ + "/v1/audio/speech" + ] }, "azure/computer-use-preview": { "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 12e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.2e-05, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -1669,15 +1797,23 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "input_cost_per_audio_token": 0.00004, - "output_cost_per_token": 0.00001, - "output_cost_per_audio_token": 0.00008, + "input_cost_per_token": 2.5e-06, + "input_cost_per_audio_token": 4e-05, + "output_cost_per_token": 1e-05, + "output_cost_per_audio_token": 8e-05, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions"], - "supported_modalities": ["text", "audio"], - "supported_output_modalities": ["text", "audio"], + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supported_modalities": [ + "text", + "audio" + ], + "supported_output_modalities": [ + "text", + "audio" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": false, @@ -1692,15 +1828,23 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "input_cost_per_audio_token": 0.00004, - "output_cost_per_token": 0.00001, - "output_cost_per_audio_token": 0.00008, + "input_cost_per_token": 2.5e-06, + "input_cost_per_audio_token": 4e-05, + "output_cost_per_token": 1e-05, + "output_cost_per_audio_token": 8e-05, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions"], - "supported_modalities": ["text", "audio"], - "supported_output_modalities": ["text", "audio"], + "supported_endpoints": [ + "/v1/chat/completions" + ], + "supported_modalities": [ + "text", + "audio" + ], + "supported_output_modalities": [ + "text", + "audio" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": false, @@ -1715,16 +1859,25 @@ "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 2e-6, - "output_cost_per_token": 8e-6, - "input_cost_per_token_batches": 1e-6, - "output_cost_per_token_batches": 4e-6, - "cache_read_input_token_cost": 0.5e-6, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, + "input_cost_per_token_batches": 1e-06, + "output_cost_per_token_batches": 4e-06, + "cache_read_input_token_cost": 5e-07, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -1735,25 +1888,34 @@ "supports_native_streaming": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 30e-3, - "search_context_size_medium": 35e-3, - "search_context_size_high": 50e-3 + "search_context_size_low": 0.03, + "search_context_size_medium": 0.035, + "search_context_size_high": 0.05 } }, "azure/gpt-4.1-2025-04-14": { "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 2e-6, - "output_cost_per_token": 8e-6, - "input_cost_per_token_batches": 1e-6, - "output_cost_per_token_batches": 4e-6, - "cache_read_input_token_cost": 0.5e-6, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, + "input_cost_per_token_batches": 1e-06, + "output_cost_per_token_batches": 4e-06, + "cache_read_input_token_cost": 5e-07, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -1764,25 +1926,34 @@ "supports_native_streaming": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 30e-3, - "search_context_size_medium": 35e-3, - "search_context_size_high": 50e-3 + "search_context_size_low": 0.03, + "search_context_size_medium": 0.035, + "search_context_size_high": 0.05 } }, "azure/gpt-4.1-mini": { "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 0.4e-6, - "output_cost_per_token": 1.6e-6, - "input_cost_per_token_batches": 0.2e-6, - "output_cost_per_token_batches": 0.8e-6, - "cache_read_input_token_cost": 0.1e-6, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 1.6e-06, + "input_cost_per_token_batches": 2e-07, + "output_cost_per_token_batches": 8e-07, + "cache_read_input_token_cost": 1e-07, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -1793,25 +1964,34 @@ "supports_native_streaming": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 25e-3, - "search_context_size_medium": 27.5e-3, - "search_context_size_high": 30e-3 + "search_context_size_low": 0.025, + "search_context_size_medium": 0.0275, + "search_context_size_high": 0.03 } }, "azure/gpt-4.1-mini-2025-04-14": { "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 0.4e-6, - "output_cost_per_token": 1.6e-6, - "input_cost_per_token_batches": 0.2e-6, - "output_cost_per_token_batches": 0.8e-6, - "cache_read_input_token_cost": 0.1e-6, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 1.6e-06, + "input_cost_per_token_batches": 2e-07, + "output_cost_per_token_batches": 8e-07, + "cache_read_input_token_cost": 1e-07, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -1822,25 +2002,34 @@ "supports_native_streaming": true, "supports_web_search": true, "search_context_cost_per_query": { - "search_context_size_low": 25e-3, - "search_context_size_medium": 27.5e-3, - "search_context_size_high": 30e-3 + "search_context_size_low": 0.025, + "search_context_size_medium": 0.0275, + "search_context_size_high": 0.03 } }, "azure/gpt-4.1-nano": { "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 0.1e-6, - "output_cost_per_token": 0.4e-6, - "input_cost_per_token_batches": 0.05e-6, - "output_cost_per_token_batches": 0.2e-6, - "cache_read_input_token_cost": 0.025e-6, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, + "input_cost_per_token_batches": 5e-08, + "output_cost_per_token_batches": 2e-07, + "cache_read_input_token_cost": 2.5e-08, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -1854,16 +2043,25 @@ "max_tokens": 32768, "max_input_tokens": 1047576, "max_output_tokens": 32768, - "input_cost_per_token": 0.1e-6, - "output_cost_per_token": 0.4e-6, - "input_cost_per_token_batches": 0.05e-6, - "output_cost_per_token_batches": 0.2e-6, - "cache_read_input_token_cost": 0.025e-6, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, + "input_cost_per_token_batches": 5e-08, + "output_cost_per_token_batches": 2e-07, + "cache_read_input_token_cost": 2.5e-08, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -1877,14 +2075,23 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 1e-5, - "output_cost_per_token": 4e-5, - "cache_read_input_token_cost": 2.5e-6, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 4e-05, + "cache_read_input_token_cost": 2.5e-06, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": false, "supports_vision": true, @@ -1897,14 +2104,23 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 1e-5, - "output_cost_per_token": 4e-5, - "cache_read_input_token_cost": 2.5e-6, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 4e-05, + "cache_read_input_token_cost": 2.5e-06, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": false, "supports_vision": true, @@ -1917,14 +2133,23 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 1.1e-6, - "output_cost_per_token": 4.4e-6, - "cache_read_input_token_cost": 2.75e-7, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 2.75e-07, "litellm_provider": "azure", "mode": "chat", - "supported_endpoints": ["/v1/chat/completions", "/v1/batch", "/v1/responses"], - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], "supports_function_calling": true, "supports_parallel_function_calling": false, "supports_vision": true, @@ -1937,12 +2162,12 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000006, - "input_cost_per_audio_token": 0.00001, - "cache_read_input_token_cost": 0.0000003, - "cache_creation_input_audio_token_cost": 0.0000003, - "output_cost_per_token": 0.0000024, - "output_cost_per_audio_token": 0.00002, + "input_cost_per_token": 6e-07, + "input_cost_per_audio_token": 1e-05, + "cache_read_input_token_cost": 3e-07, + "cache_creation_input_audio_token_cost": 3e-07, + "output_cost_per_token": 2.4e-06, + "output_cost_per_audio_token": 2e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -1956,12 +2181,12 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000066, - "input_cost_per_audio_token": 0.000011, - "cache_read_input_token_cost": 0.00000033, - "cache_creation_input_audio_token_cost": 0.00000033, - "output_cost_per_token": 0.00000264, - "output_cost_per_audio_token": 0.000022, + "input_cost_per_token": 6.6e-07, + "input_cost_per_audio_token": 1.1e-05, + "cache_read_input_token_cost": 3.3e-07, + "cache_creation_input_audio_token_cost": 3.3e-07, + "output_cost_per_token": 2.64e-06, + "output_cost_per_audio_token": 2.2e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -1975,12 +2200,12 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000066, - "input_cost_per_audio_token": 0.000011, - "cache_read_input_token_cost": 0.00000033, - "cache_creation_input_audio_token_cost": 0.00000033, - "output_cost_per_token": 0.00000264, - "output_cost_per_audio_token": 0.000022, + "input_cost_per_token": 6.6e-07, + "input_cost_per_audio_token": 1.1e-05, + "cache_read_input_token_cost": 3.3e-07, + "cache_creation_input_audio_token_cost": 3.3e-07, + "output_cost_per_token": 2.64e-06, + "output_cost_per_audio_token": 2.2e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -1994,15 +2219,21 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, - "input_cost_per_audio_token": 0.00004, - "cache_read_input_token_cost": 0.0000025, - "output_cost_per_token": 0.00002, - "output_cost_per_audio_token": 0.00008, + "input_cost_per_token": 5e-06, + "input_cost_per_audio_token": 4e-05, + "cache_read_input_token_cost": 2.5e-06, + "output_cost_per_token": 2e-05, + "output_cost_per_audio_token": 8e-05, "litellm_provider": "azure", "mode": "chat", - "supported_modalities": ["text", "audio"], - "supported_output_modalities": ["text", "audio"], + "supported_modalities": [ + "text", + "audio" + ], + "supported_output_modalities": [ + "text", + "audio" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_audio_input": true, @@ -2014,16 +2245,22 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 5.5e-6, - "input_cost_per_audio_token": 44e-6, - "cache_read_input_token_cost": 2.75e-6, - "cache_read_input_audio_token_cost": 2.5e-6, - "output_cost_per_token": 22e-6, - "output_cost_per_audio_token": 80e-6, + "input_cost_per_token": 5.5e-06, + "input_cost_per_audio_token": 4.4e-05, + "cache_read_input_token_cost": 2.75e-06, + "cache_read_input_audio_token_cost": 2.5e-06, + "output_cost_per_token": 2.2e-05, + "output_cost_per_audio_token": 8e-05, "litellm_provider": "azure", "mode": "chat", - "supported_modalities": ["text", "audio"], - "supported_output_modalities": ["text", "audio"], + "supported_modalities": [ + "text", + "audio" + ], + "supported_output_modalities": [ + "text", + "audio" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_audio_input": true, @@ -2035,16 +2272,22 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 5.5e-6, - "input_cost_per_audio_token": 44e-6, - "cache_read_input_token_cost": 2.75e-6, - "cache_read_input_audio_token_cost": 2.5e-6, - "output_cost_per_token": 22e-6, - "output_cost_per_audio_token": 80e-6, + "input_cost_per_token": 5.5e-06, + "input_cost_per_audio_token": 4.4e-05, + "cache_read_input_token_cost": 2.75e-06, + "cache_read_input_audio_token_cost": 2.5e-06, + "output_cost_per_token": 2.2e-05, + "output_cost_per_audio_token": 8e-05, "litellm_provider": "azure", "mode": "chat", - "supported_modalities": ["text", "audio"], - "supported_output_modalities": ["text", "audio"], + "supported_modalities": [ + "text", + "audio" + ], + "supported_output_modalities": [ + "text", + "audio" + ], "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_audio_input": true, @@ -2056,11 +2299,11 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, + "input_cost_per_token": 5e-06, "input_cost_per_audio_token": 0.0001, - "cache_read_input_token_cost": 0.0000025, - "cache_creation_input_audio_token_cost": 0.00002, - "output_cost_per_token": 0.00002, + "cache_read_input_token_cost": 2.5e-06, + "cache_creation_input_audio_token_cost": 2e-05, + "output_cost_per_token": 2e-05, "output_cost_per_audio_token": 0.0002, "litellm_provider": "azure", "mode": "chat", @@ -2075,11 +2318,11 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000055, + "input_cost_per_token": 5.5e-06, "input_cost_per_audio_token": 0.00011, - "cache_read_input_token_cost": 0.00000275, - "cache_creation_input_audio_token_cost": 0.000022, - "output_cost_per_token": 0.000022, + "cache_read_input_token_cost": 2.75e-06, + "cache_creation_input_audio_token_cost": 2.2e-05, + "output_cost_per_token": 2.2e-05, "output_cost_per_audio_token": 0.00022, "litellm_provider": "azure", "mode": "chat", @@ -2094,11 +2337,11 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000055, + "input_cost_per_token": 5.5e-06, "input_cost_per_audio_token": 0.00011, - "cache_read_input_token_cost": 0.00000275, - "cache_creation_input_audio_token_cost": 0.000022, - "output_cost_per_token": 0.000022, + "cache_read_input_token_cost": 2.75e-06, + "cache_creation_input_audio_token_cost": 2.2e-05, + "output_cost_per_token": 2.2e-05, "output_cost_per_audio_token": 0.00022, "litellm_provider": "azure", "mode": "chat", @@ -2113,9 +2356,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 1.1e-6, - "output_cost_per_token": 4.4e-6, - "cache_read_input_token_cost": 2.75e-7, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 2.75e-07, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2130,9 +2373,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.0000044, - "cache_read_input_token_cost": 0.00000055, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 5.5e-07, "litellm_provider": "azure", "mode": "chat", "supports_reasoning": true, @@ -2144,11 +2387,11 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.00000121, - "input_cost_per_token_batches": 0.000000605, - "output_cost_per_token": 0.00000484, - "output_cost_per_token_batches": 0.00000242, - "cache_read_input_token_cost": 0.000000605, + "input_cost_per_token": 1.21e-06, + "input_cost_per_token_batches": 6.05e-07, + "output_cost_per_token": 4.84e-06, + "output_cost_per_token_batches": 2.42e-06, + "cache_read_input_token_cost": 6.05e-07, "litellm_provider": "azure", "mode": "chat", "supports_vision": false, @@ -2160,11 +2403,11 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.00000121, - "input_cost_per_token_batches": 0.000000605, - "output_cost_per_token": 0.00000484, - "output_cost_per_token_batches": 0.00000242, - "cache_read_input_token_cost": 0.000000605, + "input_cost_per_token": 1.21e-06, + "input_cost_per_token_batches": 6.05e-07, + "output_cost_per_token": 4.84e-06, + "output_cost_per_token_batches": 2.42e-06, + "cache_read_input_token_cost": 6.05e-07, "litellm_provider": "azure", "mode": "chat", "supports_vision": false, @@ -2173,28 +2416,28 @@ "supports_tool_choice": true }, "azure/tts-1": { - "mode": "audio_speech", - "input_cost_per_character": 0.000015, + "mode": "audio_speech", + "input_cost_per_character": 1.5e-05, "litellm_provider": "azure" }, "azure/tts-1-hd": { - "mode": "audio_speech", - "input_cost_per_character": 0.000030, + "mode": "audio_speech", + "input_cost_per_character": 3e-05, "litellm_provider": "azure" }, "azure/whisper-1": { "mode": "audio_transcription", - "input_cost_per_second": 0.0001, - "output_cost_per_second": 0.0001, + "input_cost_per_second": 0.0001, + "output_cost_per_second": 0.0001, "litellm_provider": "azure" }, "azure/o3-mini": { "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.0000044, - "cache_read_input_token_cost": 0.00000055, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 5.5e-07, "litellm_provider": "azure", "mode": "chat", "supports_vision": false, @@ -2207,9 +2450,9 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.00000121, - "output_cost_per_token": 0.00000484, - "cache_read_input_token_cost": 0.000000605, + "input_cost_per_token": 1.21e-06, + "output_cost_per_token": 4.84e-06, + "cache_read_input_token_cost": 6.05e-07, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2222,9 +2465,9 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 1.1e-6, - "output_cost_per_token": 4.4e-6, - "cache_read_input_token_cost": 0.55e-6, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, + "cache_read_input_token_cost": 5.5e-07, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2237,11 +2480,11 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.00000121, - "input_cost_per_token_batches": 0.000000605, - "output_cost_per_token": 0.00000484, - "output_cost_per_token_batches": 0.00000242, - "cache_read_input_token_cost": 0.000000605, + "input_cost_per_token": 1.21e-06, + "input_cost_per_token_batches": 6.05e-07, + "output_cost_per_token": 4.84e-06, + "output_cost_per_token_batches": 2.42e-06, + "cache_read_input_token_cost": 6.05e-07, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2253,11 +2496,11 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.00000121, - "input_cost_per_token_batches": 0.000000605, - "output_cost_per_token": 0.00000484, - "output_cost_per_token_batches": 0.00000242, - "cache_read_input_token_cost": 0.000000605, + "input_cost_per_token": 1.21e-06, + "input_cost_per_token_batches": 6.05e-07, + "output_cost_per_token": 4.84e-06, + "output_cost_per_token_batches": 2.42e-06, + "cache_read_input_token_cost": 6.05e-07, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2269,9 +2512,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2285,9 +2528,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2301,9 +2544,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.0000165, - "output_cost_per_token": 0.000066, - "cache_read_input_token_cost": 0.00000825, + "input_cost_per_token": 1.65e-05, + "output_cost_per_token": 6.6e-05, + "cache_read_input_token_cost": 8.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2316,9 +2559,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.0000165, - "output_cost_per_token": 0.000066, - "cache_read_input_token_cost": 0.00000825, + "input_cost_per_token": 1.65e-05, + "output_cost_per_token": 6.6e-05, + "cache_read_input_token_cost": 8.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2331,9 +2574,9 @@ "max_tokens": 32768, "max_input_tokens": 128000, "max_output_tokens": 32768, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2346,9 +2589,9 @@ "max_tokens": 32768, "max_input_tokens": 128000, "max_output_tokens": 32768, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "azure", "mode": "chat", "supports_pdf_input": true, @@ -2362,9 +2605,9 @@ "max_tokens": 32768, "max_input_tokens": 128000, "max_output_tokens": 32768, - "input_cost_per_token": 0.0000165, - "output_cost_per_token": 0.000066, - "cache_read_input_token_cost": 0.00000825, + "input_cost_per_token": 1.65e-05, + "output_cost_per_token": 6.6e-05, + "cache_read_input_token_cost": 8.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2376,9 +2619,9 @@ "max_tokens": 32768, "max_input_tokens": 128000, "max_output_tokens": 32768, - "input_cost_per_token": 0.0000165, - "output_cost_per_token": 0.000066, - "cache_read_input_token_cost": 0.00000825, + "input_cost_per_token": 1.65e-05, + "output_cost_per_token": 6.6e-05, + "cache_read_input_token_cost": 8.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2390,11 +2633,11 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.000075, + "input_cost_per_token": 7.5e-05, "output_cost_per_token": 0.00015, - "input_cost_per_token_batches": 0.0000375, - "output_cost_per_token_batches": 0.000075, - "cache_read_input_token_cost": 0.0000375, + "input_cost_per_token_batches": 3.75e-05, + "output_cost_per_token_batches": 7.5e-05, + "cache_read_input_token_cost": 3.75e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2409,9 +2652,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2425,9 +2668,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2441,9 +2684,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2457,9 +2700,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2473,9 +2716,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000275, - "output_cost_per_token": 0.000011, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.75e-06, + "output_cost_per_token": 1.1e-05, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2489,9 +2732,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000275, - "cache_creation_input_token_cost": 0.00000138, - "output_cost_per_token": 0.000011, + "input_cost_per_token": 2.75e-06, + "cache_creation_input_token_cost": 1.38e-06, + "output_cost_per_token": 1.1e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2504,9 +2747,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000275, - "cache_creation_input_token_cost": 0.00000138, - "output_cost_per_token": 0.000011, + "input_cost_per_token": 2.75e-06, + "cache_creation_input_token_cost": 1.38e-06, + "output_cost_per_token": 1.1e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2519,8 +2762,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2533,9 +2776,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000010, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2550,9 +2793,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000275, - "output_cost_per_token": 0.000011, - "cache_read_input_token_cost": 0.000001375, + "input_cost_per_token": 2.75e-06, + "output_cost_per_token": 1.1e-05, + "cache_read_input_token_cost": 1.375e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2566,9 +2809,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000275, - "output_cost_per_token": 0.000011, - "cache_read_input_token_cost": 0.000001375, + "input_cost_per_token": 2.75e-06, + "output_cost_per_token": 1.1e-05, + "cache_read_input_token_cost": 1.375e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2582,9 +2825,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000010, - "cache_read_input_token_cost": 0.00000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, + "cache_read_input_token_cost": 1.25e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2598,8 +2841,8 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000060, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2612,9 +2855,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.000000165, - "output_cost_per_token": 0.00000066, - "cache_read_input_token_cost": 0.000000075, + "input_cost_per_token": 1.65e-07, + "output_cost_per_token": 6.6e-07, + "cache_read_input_token_cost": 7.5e-08, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2628,9 +2871,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.000000165, - "output_cost_per_token": 0.00000066, - "cache_read_input_token_cost": 0.000000075, + "input_cost_per_token": 1.65e-07, + "output_cost_per_token": 6.6e-07, + "cache_read_input_token_cost": 7.5e-08, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2644,9 +2887,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.000000165, - "output_cost_per_token": 0.00000066, - "cache_read_input_token_cost": 0.000000083, + "input_cost_per_token": 1.65e-07, + "output_cost_per_token": 6.6e-07, + "cache_read_input_token_cost": 8.3e-08, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2660,9 +2903,9 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 0.000000165, - "output_cost_per_token": 0.00000066, - "cache_read_input_token_cost": 0.000000083, + "input_cost_per_token": 1.65e-07, + "output_cost_per_token": 6.6e-07, + "cache_read_input_token_cost": 8.3e-08, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2676,8 +2919,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2689,8 +2932,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2701,8 +2944,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2713,8 +2956,8 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.00006, + "input_cost_per_token": 3e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2724,7 +2967,7 @@ "max_tokens": 4096, "max_input_tokens": 32768, "max_output_tokens": 4096, - "input_cost_per_token": 0.00006, + "input_cost_per_token": 6e-05, "output_cost_per_token": 0.00012, "litellm_provider": "azure", "mode": "chat", @@ -2734,7 +2977,7 @@ "max_tokens": 4096, "max_input_tokens": 32768, "max_output_tokens": 4096, - "input_cost_per_token": 0.00006, + "input_cost_per_token": 6e-05, "output_cost_per_token": 0.00012, "litellm_provider": "azure", "mode": "chat", @@ -2744,8 +2987,8 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.00006, + "input_cost_per_token": 3e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2755,9 +2998,9 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, - "litellm_provider": "azure", + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, + "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -2767,9 +3010,9 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, - "litellm_provider": "azure", + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, + "litellm_provider": "azure", "mode": "chat", "supports_vision": true, "supports_tool_choice": true @@ -2778,8 +3021,8 @@ "max_tokens": 4096, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000004, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 4e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2789,8 +3032,8 @@ "max_tokens": 4096, "max_input_tokens": 16384, "max_output_tokens": 4096, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2802,8 +3045,8 @@ "max_tokens": 4097, "max_input_tokens": 4097, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2815,8 +3058,8 @@ "max_tokens": 4097, "max_input_tokens": 4097, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2828,8 +3071,8 @@ "max_tokens": 4096, "max_input_tokens": 16384, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2841,8 +3084,8 @@ "max_tokens": 4096, "max_input_tokens": 16384, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2854,8 +3097,8 @@ "max_tokens": 4096, "max_input_tokens": 16385, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000004, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 4e-06, "litellm_provider": "azure", "mode": "chat", "supports_tool_choice": true @@ -2864,8 +3107,8 @@ "max_tokens": 4096, "max_input_tokens": 4097, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2875,8 +3118,8 @@ "max_tokens": 4096, "max_input_tokens": 4097, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true, @@ -2885,32 +3128,32 @@ "azure/gpt-3.5-turbo-instruct-0914": { "max_tokens": 4097, "max_input_tokens": 4097, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "azure_text", "mode": "completion" }, "azure/gpt-35-turbo-instruct": { "max_tokens": 4097, "max_input_tokens": 4097, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "azure_text", "mode": "completion" }, "azure/gpt-35-turbo-instruct-0914": { "max_tokens": 4097, "max_input_tokens": 4097, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "azure_text", "mode": "completion" }, "azure/mistral-large-latest": { "max_tokens": 32000, "max_input_tokens": 32000, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true @@ -2918,18 +3161,18 @@ "azure/mistral-large-2402": { "max_tokens": 32000, "max_input_tokens": 32000, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true }, "azure/command-r-plus": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "azure", "mode": "chat", "supports_function_calling": true @@ -2937,153 +3180,173 @@ "azure/ada": { "max_tokens": 8191, "max_input_tokens": 8191, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "azure", "mode": "embedding" }, "azure/text-embedding-ada-002": { "max_tokens": 8191, "max_input_tokens": 8191, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "azure", "mode": "embedding" }, "azure/text-embedding-3-large": { "max_tokens": 8191, "max_input_tokens": 8191, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 0.0, "litellm_provider": "azure", "mode": "embedding" }, "azure/text-embedding-3-small": { "max_tokens": 8191, "max_input_tokens": 8191, - "input_cost_per_token": 0.00000002, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 2e-08, + "output_cost_per_token": 0.0, "litellm_provider": "azure", "mode": "embedding" }, "azure/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 4.0054321e-8, + "input_cost_per_pixel": 4.0054321e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/low/1024-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.0490417e-8, + "input_cost_per_pixel": 1.0490417e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/medium/1024-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 4.0054321e-8, + "input_cost_per_pixel": 4.0054321e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/high/1024-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.59263611e-7, + "input_cost_per_pixel": 1.59263611e-07, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/low/1024-x-1536/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.0172526e-8, + "input_cost_per_pixel": 1.0172526e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/medium/1024-x-1536/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 4.0054321e-8, + "input_cost_per_pixel": 4.0054321e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/high/1024-x-1536/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.58945719e-7, + "input_cost_per_pixel": 1.58945719e-07, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/low/1536-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.0172526e-8, + "input_cost_per_pixel": 1.0172526e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/medium/1536-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 4.0054321e-8, + "input_cost_per_pixel": 4.0054321e-08, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] + "supported_endpoints": [ + "/v1/images/generations" + ] }, "azure/high/1536-x-1024/gpt-image-1": { "mode": "image_generation", - "input_cost_per_pixel": 1.58945719e-7, + "input_cost_per_pixel": 1.58945719e-07, "output_cost_per_pixel": 0.0, "litellm_provider": "azure", - "supported_endpoints": ["/v1/images/generations"] - }, + "supported_endpoints": [ + "/v1/images/generations" + ] + }, "azure/standard/1024-x-1024/dall-e-3": { - "input_cost_per_pixel": 0.0000000381469, + "input_cost_per_pixel": 3.81469e-08, "output_cost_per_token": 0.0, - "litellm_provider": "azure", + "litellm_provider": "azure", "mode": "image_generation" }, "azure/hd/1024-x-1024/dall-e-3": { - "input_cost_per_pixel": 0.00000007629, + "input_cost_per_pixel": 7.629e-08, "output_cost_per_token": 0.0, - "litellm_provider": "azure", + "litellm_provider": "azure", "mode": "image_generation" }, "azure/standard/1024-x-1792/dall-e-3": { - "input_cost_per_pixel": 0.00000004359, + "input_cost_per_pixel": 4.359e-08, "output_cost_per_token": 0.0, - "litellm_provider": "azure", + "litellm_provider": "azure", "mode": "image_generation" }, "azure/standard/1792-x-1024/dall-e-3": { - "input_cost_per_pixel": 0.00000004359, + "input_cost_per_pixel": 4.359e-08, "output_cost_per_token": 0.0, - "litellm_provider": "azure", + "litellm_provider": "azure", "mode": "image_generation" }, "azure/hd/1024-x-1792/dall-e-3": { - "input_cost_per_pixel": 0.00000006539, + "input_cost_per_pixel": 6.539e-08, "output_cost_per_token": 0.0, - "litellm_provider": "azure", + "litellm_provider": "azure", "mode": "image_generation" }, "azure/hd/1792-x-1024/dall-e-3": { - "input_cost_per_pixel": 0.00000006539, + "input_cost_per_pixel": 6.539e-08, "output_cost_per_token": 0.0, - "litellm_provider": "azure", + "litellm_provider": "azure", "mode": "image_generation" }, "azure/standard/1024-x-1024/dall-e-2": { "input_cost_per_pixel": 0.0, "output_cost_per_token": 0.0, - "litellm_provider": "azure", + "litellm_provider": "azure", "mode": "image_generation" }, "azure_ai/deepseek-r1": { "max_tokens": 8192, "max_input_tokens": 128000, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000135, - "output_cost_per_token": 0.0000054, + "input_cost_per_token": 1.35e-06, + "output_cost_per_token": 5.4e-06, "litellm_provider": "azure_ai", "mode": "chat", "supports_tool_choice": true, @@ -3094,8 +3357,8 @@ "max_tokens": 8192, "max_input_tokens": 128000, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000114, - "output_cost_per_token": 0.00000456, + "input_cost_per_token": 1.14e-06, + "output_cost_per_token": 4.56e-06, "litellm_provider": "azure_ai", "mode": "chat", "supports_tool_choice": true, @@ -3105,8 +3368,8 @@ "max_tokens": 8192, "max_input_tokens": 128000, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000114, - "output_cost_per_token": 0.00000456, + "input_cost_per_token": 1.14e-06, + "output_cost_per_token": 4.56e-06, "litellm_provider": "azure_ai", "mode": "chat", "supports_function_calling": true, @@ -3117,8 +3380,8 @@ "max_tokens": 4096, "max_input_tokens": 70000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000007, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_tool_choice": true @@ -3127,8 +3390,8 @@ "max_tokens": 4096, "max_input_tokens": 131072, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_function_calling": true, @@ -3138,8 +3401,8 @@ "max_tokens": 8191, "max_input_tokens": 131072, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000004, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 2e-06, "litellm_provider": "azure_ai", "mode": "chat", "supports_function_calling": true, @@ -3150,8 +3413,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000004, - "output_cost_per_token": 0.000012, + "input_cost_per_token": 4e-06, + "output_cost_per_token": 1.2e-05, "litellm_provider": "azure_ai", "mode": "chat", "supports_function_calling": true, @@ -3161,8 +3424,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "azure_ai", "supports_function_calling": true, "mode": "chat", @@ -3172,8 +3435,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "azure_ai", "mode": "chat", "supports_function_calling": true, @@ -3184,8 +3447,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "azure_ai", "supports_function_calling": true, "mode": "chat", @@ -3196,8 +3459,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "azure_ai", "supports_function_calling": true, "mode": "chat", @@ -3208,20 +3471,20 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000004, - "output_cost_per_token": 0.00000004, + "input_cost_per_token": 4e-08, + "output_cost_per_token": 4e-08, "litellm_provider": "azure_ai", "supports_function_calling": true, "mode": "chat", "source": "https://azuremarketplace.microsoft.com/en/marketplace/apps/000-000.ministral-3b-2410-offer?tab=Overview", "supports_tool_choice": true - }, + }, "azure_ai/Llama-3.2-11B-Vision-Instruct": { "max_tokens": 2048, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000037, - "output_cost_per_token": 0.00000037, + "input_cost_per_token": 3.7e-07, + "output_cost_per_token": 3.7e-07, "litellm_provider": "azure_ai", "supports_function_calling": true, "supports_vision": true, @@ -3233,8 +3496,8 @@ "max_tokens": 2048, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000071, - "output_cost_per_token": 0.00000071, + "input_cost_per_token": 7.1e-07, + "output_cost_per_token": 7.1e-07, "litellm_provider": "azure_ai", "supports_function_calling": true, "mode": "chat", @@ -3245,8 +3508,8 @@ "max_tokens": 16384, "max_input_tokens": 10000000, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.00000078, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 7.8e-07, "litellm_provider": "azure_ai", "supports_function_calling": true, "supports_vision": true, @@ -3258,8 +3521,8 @@ "max_tokens": 16384, "max_input_tokens": 1000000, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000141, - "output_cost_per_token": 0.00000035, + "input_cost_per_token": 1.41e-06, + "output_cost_per_token": 3.5e-07, "litellm_provider": "azure_ai", "supports_function_calling": true, "supports_vision": true, @@ -3271,8 +3534,8 @@ "max_tokens": 2048, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000204, - "output_cost_per_token": 0.00000204, + "input_cost_per_token": 2.04e-06, + "output_cost_per_token": 2.04e-06, "litellm_provider": "azure_ai", "supports_function_calling": true, "supports_vision": true, @@ -3284,8 +3547,8 @@ "max_tokens": 2048, "max_input_tokens": 8192, "max_output_tokens": 2048, - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.00000037, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 3.7e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_tool_choice": true @@ -3294,41 +3557,41 @@ "max_tokens": 2048, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.00000061, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 6.1e-07, "litellm_provider": "azure_ai", "mode": "chat", - "source":"https://azuremarketplace.microsoft.com/en-us/marketplace/apps/metagenai.meta-llama-3-1-8b-instruct-offer?tab=PlansAndPrice", + "source": "https://azuremarketplace.microsoft.com/en-us/marketplace/apps/metagenai.meta-llama-3-1-8b-instruct-offer?tab=PlansAndPrice", "supports_tool_choice": true }, "azure_ai/Meta-Llama-3.1-70B-Instruct": { "max_tokens": 2048, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000268, - "output_cost_per_token": 0.00000354, + "input_cost_per_token": 2.68e-06, + "output_cost_per_token": 3.54e-06, "litellm_provider": "azure_ai", "mode": "chat", - "source":"https://azuremarketplace.microsoft.com/en-us/marketplace/apps/metagenai.meta-llama-3-1-70b-instruct-offer?tab=PlansAndPrice", + "source": "https://azuremarketplace.microsoft.com/en-us/marketplace/apps/metagenai.meta-llama-3-1-70b-instruct-offer?tab=PlansAndPrice", "supports_tool_choice": true }, "azure_ai/Meta-Llama-3.1-405B-Instruct": { "max_tokens": 2048, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000533, - "output_cost_per_token": 0.000016, + "input_cost_per_token": 5.33e-06, + "output_cost_per_token": 1.6e-05, "litellm_provider": "azure_ai", "mode": "chat", - "source":"https://azuremarketplace.microsoft.com/en-us/marketplace/apps/metagenai.meta-llama-3-1-405b-instruct-offer?tab=PlansAndPrice", + "source": "https://azuremarketplace.microsoft.com/en-us/marketplace/apps/metagenai.meta-llama-3-1-405b-instruct-offer?tab=PlansAndPrice", "supports_tool_choice": true }, "azure_ai/Phi-4-mini-instruct": { "max_tokens": 4096, "max_input_tokens": 131072, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000075, - "output_cost_per_token": 0.0000003, + "input_cost_per_token": 7.5e-08, + "output_cost_per_token": 3e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_function_calling": true, @@ -3338,9 +3601,9 @@ "max_tokens": 4096, "max_input_tokens": 131072, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000008, - "input_cost_per_audio_token": 0.000004, - "output_cost_per_token": 0.00000032, + "input_cost_per_token": 8e-08, + "input_cost_per_audio_token": 4e-06, + "output_cost_per_token": 3.2e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_audio_input": true, @@ -3352,8 +3615,8 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3365,8 +3628,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000052, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 5.2e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3377,8 +3640,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000052, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 5.2e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": true, @@ -3389,8 +3652,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000016, - "output_cost_per_token": 0.00000064, + "input_cost_per_token": 1.6e-07, + "output_cost_per_token": 6.4e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3401,8 +3664,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000052, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 5.2e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3413,8 +3676,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000052, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 5.2e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3425,8 +3688,8 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3437,8 +3700,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3449,8 +3712,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000017, - "output_cost_per_token": 0.00000068, + "input_cost_per_token": 1.7e-07, + "output_cost_per_token": 6.8e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3461,8 +3724,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000017, - "output_cost_per_token": 0.00000068, + "input_cost_per_token": 1.7e-07, + "output_cost_per_token": 6.8e-07, "litellm_provider": "azure_ai", "mode": "chat", "supports_vision": false, @@ -3495,43 +3758,48 @@ "max_tokens": 512, "max_input_tokens": 512, "output_vector_size": 1024, - "input_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0.0, "litellm_provider": "azure_ai", "mode": "embedding", "supports_embedding_image_input": true, - "source":"https://azuremarketplace.microsoft.com/en-us/marketplace/apps/cohere.cohere-embed-v3-english-offer?tab=PlansAndPrice" + "source": "https://azuremarketplace.microsoft.com/en-us/marketplace/apps/cohere.cohere-embed-v3-english-offer?tab=PlansAndPrice" }, "azure_ai/Cohere-embed-v3-multilingual": { "max_tokens": 512, "max_input_tokens": 512, "output_vector_size": 1024, - "input_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0.0, "litellm_provider": "azure_ai", "mode": "embedding", "supports_embedding_image_input": true, - "source":"https://azuremarketplace.microsoft.com/en-us/marketplace/apps/cohere.cohere-embed-v3-english-offer?tab=PlansAndPrice" + "source": "https://azuremarketplace.microsoft.com/en-us/marketplace/apps/cohere.cohere-embed-v3-english-offer?tab=PlansAndPrice" }, "azure_ai/embed-v-4-0": { "max_tokens": 128000, "max_input_tokens": 128000, "output_vector_size": 3072, - "input_cost_per_token": 0.00000012, + "input_cost_per_token": 1.2e-07, "output_cost_per_token": 0.0, "litellm_provider": "azure_ai", "mode": "embedding", "supports_embedding_image_input": true, - "supported_endpoints": ["/v1/embeddings"], - "supported_modalities": ["text", "image"], - "source":"https://azuremarketplace.microsoft.com/pt-br/marketplace/apps/cohere.cohere-embed-4-offer?tab=PlansAndPrice" + "supported_endpoints": [ + "/v1/embeddings" + ], + "supported_modalities": [ + "text", + "image" + ], + "source": "https://azuremarketplace.microsoft.com/pt-br/marketplace/apps/cohere.cohere-embed-4-offer?tab=PlansAndPrice" }, "babbage-002": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000004, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "text-completion-openai", "mode": "completion" }, @@ -3539,17 +3807,17 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 4096, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "text-completion-openai", "mode": "completion" - }, + }, "gpt-3.5-turbo-instruct": { "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "text-completion-openai", "mode": "completion" }, @@ -3557,18 +3825,17 @@ "max_tokens": 4097, "max_input_tokens": 8192, "max_output_tokens": 4097, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "text-completion-openai", "mode": "completion" - }, "claude-instant-1": { "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000163, - "output_cost_per_token": 0.00000551, + "input_cost_per_token": 1.63e-06, + "output_cost_per_token": 5.51e-06, "litellm_provider": "anthropic", "mode": "chat" }, @@ -3576,8 +3843,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 2.5e-07, "litellm_provider": "mistral", "mode": "chat", "supports_assistant_prefill": true, @@ -3587,8 +3854,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000003, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 3e-07, "litellm_provider": "mistral", "supports_function_calling": true, "mode": "chat", @@ -3599,8 +3866,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000003, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 3e-07, "litellm_provider": "mistral", "supports_function_calling": true, "mode": "chat", @@ -3611,8 +3878,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000027, - "output_cost_per_token": 0.0000081, + "input_cost_per_token": 2.7e-06, + "output_cost_per_token": 8.1e-06, "litellm_provider": "mistral", "mode": "chat", "supports_assistant_prefill": true, @@ -3622,8 +3889,8 @@ "max_tokens": 8191, "max_input_tokens": 131072, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000004, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 2e-06, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3634,8 +3901,8 @@ "max_tokens": 8191, "max_input_tokens": 131072, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000004, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 2e-06, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3646,8 +3913,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000027, - "output_cost_per_token": 0.0000081, + "input_cost_per_token": 2.7e-06, + "output_cost_per_token": 8.1e-06, "litellm_provider": "mistral", "mode": "chat", "supports_assistant_prefill": true, @@ -3657,8 +3924,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3669,8 +3936,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3681,8 +3948,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000004, - "output_cost_per_token": 0.000012, + "input_cost_per_token": 4e-06, + "output_cost_per_token": 1.2e-05, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3693,8 +3960,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000009, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 9e-06, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3705,8 +3972,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3718,8 +3985,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3731,8 +3998,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3744,8 +4011,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 2.5e-07, "litellm_provider": "mistral", "mode": "chat", "supports_assistant_prefill": true, @@ -3755,8 +4022,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000007, - "output_cost_per_token": 0.0000007, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3767,8 +4034,8 @@ "max_tokens": 8191, "max_input_tokens": 65336, "max_output_tokens": 8191, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "mistral", "mode": "chat", "supports_function_calling": true, @@ -3779,8 +4046,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "mistral", "mode": "chat", "supports_assistant_prefill": true, @@ -3790,8 +4057,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "mistral", "mode": "chat", "supports_assistant_prefill": true, @@ -3801,8 +4068,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000003, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 3e-07, "litellm_provider": "mistral", "mode": "chat", "source": "https://mistral.ai/technology/", @@ -3813,8 +4080,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000003, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 3e-07, "litellm_provider": "mistral", "mode": "chat", "source": "https://mistral.ai/technology/", @@ -3825,8 +4092,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 2.5e-07, "litellm_provider": "mistral", "mode": "chat", "source": "https://mistral.ai/technology/", @@ -3837,8 +4104,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 2.5e-07, "litellm_provider": "mistral", "mode": "chat", "source": "https://mistral.ai/technology/", @@ -3849,8 +4116,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000003, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 3e-07, "litellm_provider": "mistral", "mode": "chat", "source": "https://mistral.ai/news/devstral", @@ -3861,7 +4128,7 @@ "mistral/mistral-embed": { "max_tokens": 8192, "max_input_tokens": 8192, - "input_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, "litellm_provider": "mistral", "mode": "embedding" }, @@ -3869,12 +4136,12 @@ "max_tokens": 8192, "max_input_tokens": 65536, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000055, - "input_cost_per_token_cache_hit": 0.00000014, - "output_cost_per_token": 0.00000219, + "input_cost_per_token": 5.5e-07, + "input_cost_per_token_cache_hit": 1.4e-07, + "output_cost_per_token": 2.19e-06, "litellm_provider": "deepseek", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_assistant_prefill": true, "supports_tool_choice": true, "supports_reasoning": true, @@ -3884,14 +4151,14 @@ "max_tokens": 8192, "max_input_tokens": 65536, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000027, - "input_cost_per_token_cache_hit": 0.00000007, - "cache_read_input_token_cost": 0.00000007, + "input_cost_per_token": 2.7e-07, + "input_cost_per_token_cache_hit": 7e-08, + "cache_read_input_token_cost": 7e-08, "cache_creation_input_token_cost": 0.0, - "output_cost_per_token": 0.0000011, + "output_cost_per_token": 1.1e-06, "litellm_provider": "deepseek", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_assistant_prefill": true, "supports_tool_choice": true, "supports_prompt_caching": true @@ -3900,8 +4167,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000000, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "codestral", "mode": "chat", "source": "https://docs.mistral.ai/capabilities/code_generation/", @@ -3912,8 +4179,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000000, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "codestral", "mode": "chat", "source": "https://docs.mistral.ai/capabilities/code_generation/", @@ -3924,8 +4191,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000000, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "text-completion-codestral", "mode": "completion", "source": "https://docs.mistral.ai/capabilities/code_generation/" @@ -3934,8 +4201,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000000, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "text-completion-codestral", "mode": "completion", "source": "https://docs.mistral.ai/capabilities/code_generation/" @@ -3944,8 +4211,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -3956,9 +4223,9 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 32768, - "input_cost_per_token": 0.000002, - "input_cost_per_image": 0.000002, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2e-06, + "input_cost_per_image": 2e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -3969,9 +4236,9 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 32768, - "input_cost_per_token": 0.000002, - "input_cost_per_image": 0.000002, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2e-06, + "input_cost_per_image": 2e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -3982,9 +4249,9 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 32768, - "input_cost_per_token": 0.000002, - "input_cost_per_image": 0.000002, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2e-06, + "input_cost_per_image": 2e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -3995,8 +4262,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4008,8 +4275,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4021,8 +4288,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000025, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 2.5e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4034,8 +4301,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000025, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 2.5e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4047,8 +4314,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4061,8 +4328,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.000004, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 4e-06, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4075,8 +4342,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.000004, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 4e-06, "litellm_provider": "xai", "mode": "chat", "supports_reasoning": true, @@ -4089,9 +4356,9 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.000005, - "input_cost_per_image": 0.000005, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 5e-06, + "input_cost_per_image": 5e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4102,8 +4369,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4113,8 +4380,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4124,8 +4391,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, @@ -4135,12 +4402,12 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000014, - "input_cost_per_token_cache_hit": 0.000000014, - "output_cost_per_token": 0.00000028, + "input_cost_per_token": 1.4e-07, + "input_cost_per_token_cache_hit": 1.4e-08, + "output_cost_per_token": 2.8e-07, "litellm_provider": "deepseek", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_assistant_prefill": true, "supports_tool_choice": true, "supports_prompt_caching": true @@ -4194,8 +4461,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000080, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 8e-07, "litellm_provider": "groq", "mode": "chat", "supports_function_calling": true, @@ -4332,8 +4599,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000059, - "output_cost_per_token": 0.00000079, + "input_cost_per_token": 5.9e-07, + "output_cost_per_token": 7.9e-07, "litellm_provider": "groq", "mode": "chat", "supports_function_calling": true, @@ -4480,8 +4747,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 1e-07, "litellm_provider": "cerebras", "mode": "chat", "supports_function_calling": true, @@ -4491,8 +4758,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "cerebras", "mode": "chat", "supports_function_calling": true, @@ -4502,8 +4769,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.00000085, - "output_cost_per_token": 0.0000012, + "input_cost_per_token": 8.5e-07, + "output_cost_per_token": 1.2e-06, "litellm_provider": "cerebras", "mode": "chat", "supports_function_calling": true, @@ -4513,8 +4780,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 1e-07, "litellm_provider": "friendliai", "mode": "chat", "supports_function_calling": true, @@ -4527,8 +4794,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "friendliai", "mode": "chat", "supports_function_calling": true, @@ -4541,8 +4808,8 @@ "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000000163, - "output_cost_per_token": 0.000000551, + "input_cost_per_token": 1.63e-07, + "output_cost_per_token": 5.51e-07, "litellm_provider": "anthropic", "mode": "chat", "supports_tool_choice": true @@ -4551,8 +4818,8 @@ "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "anthropic", "mode": "chat" }, @@ -4560,8 +4827,8 @@ "max_tokens": 8191, "max_input_tokens": 200000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "anthropic", "mode": "chat", "supports_tool_choice": true @@ -4570,10 +4837,10 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, - "cache_creation_input_token_cost": 0.0000003, - "cache_read_input_token_cost": 0.00000003, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, + "cache_creation_input_token_cost": 3e-07, + "cache_read_input_token_cost": 3e-08, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4589,14 +4856,14 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.000004, - "cache_creation_input_token_cost": 0.000001, - "cache_read_input_token_cost": 0.00000008, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 4e-06, + "cache_creation_input_token_cost": 1e-06, + "cache_read_input_token_cost": 8e-08, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, "litellm_provider": "anthropic", "mode": "chat", @@ -4615,14 +4882,14 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000005, - "cache_creation_input_token_cost": 0.00000125, - "cache_read_input_token_cost": 0.0000001, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 5e-06, + "cache_creation_input_token_cost": 1.25e-06, + "cache_read_input_token_cost": 1e-07, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, "litellm_provider": "anthropic", "mode": "chat", @@ -4641,10 +4908,10 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000075, - "cache_creation_input_token_cost": 0.00001875, - "cache_read_input_token_cost": 0.0000015, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, + "cache_creation_input_token_cost": 1.875e-05, + "cache_read_input_token_cost": 1.5e-06, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4660,10 +4927,10 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000075, - "cache_creation_input_token_cost": 0.00001875, - "cache_read_input_token_cost": 0.0000015, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, + "cache_creation_input_token_cost": 1.875e-05, + "cache_read_input_token_cost": 1.5e-06, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4679,8 +4946,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4697,14 +4964,14 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, "litellm_provider": "anthropic", "mode": "chat", @@ -4723,10 +4990,10 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4743,15 +5010,15 @@ "max_tokens": 32000, "max_input_tokens": 200000, "max_output_tokens": 32000, - "input_cost_per_token": 15e-6, - "output_cost_per_token": 75e-6, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 18.75e-6, - "cache_read_input_token_cost": 1.5e-6, + "cache_creation_input_token_cost": 1.875e-05, + "cache_read_input_token_cost": 1.5e-06, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4769,15 +5036,15 @@ "max_tokens": 64000, "max_input_tokens": 200000, "max_output_tokens": 64000, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 15e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 3.75e-6, - "cache_read_input_token_cost": 0.3e-6, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4796,15 +5063,15 @@ "max_tokens": 128000, "max_input_tokens": 200000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4823,14 +5090,14 @@ "max_tokens": 128000, "max_input_tokens": 200000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, "litellm_provider": "anthropic", "mode": "chat", @@ -4851,14 +5118,14 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, "litellm_provider": "anthropic", "mode": "chat", @@ -4877,8 +5144,8 @@ "max_tokens": 2048, "max_input_tokens": 8192, "max_output_tokens": 2048, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -4887,8 +5154,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -4897,8 +5164,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -4907,10 +5174,10 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -4919,10 +5186,10 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -4931,8 +5198,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.000028, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 2.8e-05, "litellm_provider": "vertex_ai-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -4941,8 +5208,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.000028, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 2.8e-05, "litellm_provider": "vertex_ai-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -4951,10 +5218,10 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -4964,10 +5231,10 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -4977,10 +5244,10 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -4991,10 +5258,10 @@ "max_tokens": 8192, "max_input_tokens": 32000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5004,10 +5271,10 @@ "max_tokens": 8192, "max_input_tokens": 32000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5017,10 +5284,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5030,10 +5297,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -5042,10 +5309,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -5054,10 +5321,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -5066,10 +5333,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -5078,8 +5345,8 @@ "max_tokens": 64, "max_input_tokens": 2048, "max_output_tokens": 64, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -5088,8 +5355,8 @@ "max_tokens": 64, "max_input_tokens": 2048, "max_output_tokens": 64, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -5098,8 +5365,8 @@ "max_tokens": 64, "max_input_tokens": 2048, "max_output_tokens": 64, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -5108,8 +5375,8 @@ "max_tokens": 64, "max_input_tokens": 2048, "max_output_tokens": 64, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "vertex_ai-code-text-models", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -5118,10 +5385,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5131,10 +5398,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5144,10 +5411,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5157,10 +5424,10 @@ "max_tokens": 1024, "max_input_tokens": 6144, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5170,10 +5437,10 @@ "max_tokens": 8192, "max_input_tokens": 32000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5183,10 +5450,10 @@ "max_tokens": 8192, "max_input_tokens": 32000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, - "input_cost_per_character": 0.00000025, - "output_cost_per_character": 0.0000005, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, + "input_cost_per_character": 2.5e-07, + "output_cost_per_character": 5e-07, "litellm_provider": "vertex_ai-code-chat-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5201,8 +5468,13 @@ "supports_function_calling": false, "source": "https://llama.developer.meta.com/docs/models", "supports_tool_choice": false, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ] }, "meta_llama/Llama-4-Maverick-17B-128E-Instruct-FP8": { "max_tokens": 128000, @@ -5213,8 +5485,13 @@ "supports_function_calling": false, "source": "https://llama.developer.meta.com/docs/models", "supports_tool_choice": false, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ] }, "meta_llama/Llama-3.3-70B-Instruct": { "max_tokens": 128000, @@ -5225,8 +5502,12 @@ "supports_function_calling": false, "source": "https://llama.developer.meta.com/docs/models", "supports_tool_choice": false, - "supported_modalities": ["text"], - "supported_output_modalities": ["text"] + "supported_modalities": [ + "text" + ], + "supported_output_modalities": [ + "text" + ] }, "meta_llama/Llama-3.3-8B-Instruct": { "max_tokens": 128000, @@ -5237,8 +5518,12 @@ "supports_function_calling": false, "source": "https://llama.developer.meta.com/docs/models", "supports_tool_choice": false, - "supported_modalities": ["text"], - "supported_output_modalities": ["text"] + "supported_modalities": [ + "text" + ], + "supported_output_modalities": [ + "text" + ] }, "gemini-pro": { "max_tokens": 8192, @@ -5246,48 +5531,51 @@ "max_output_tokens": 8192, "input_cost_per_image": 0.0025, "input_cost_per_video_per_second": 0.002, - "input_cost_per_token": 0.0000005, - "input_cost_per_character": 0.000000125, - "output_cost_per_token": 0.0000015, - "output_cost_per_character": 0.000000375, + "input_cost_per_token": 5e-07, + "input_cost_per_character": 1.25e-07, + "output_cost_per_token": 1.5e-06, + "output_cost_per_character": 3.75e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": true, + "supports_parallel_function_calling": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing", "supports_tool_choice": true }, - "gemini-1.0-pro": { + "gemini-1.0-pro": { "max_tokens": 8192, "max_input_tokens": 32760, "max_output_tokens": 8192, "input_cost_per_image": 0.0025, "input_cost_per_video_per_second": 0.002, - "input_cost_per_token": 0.0000005, - "input_cost_per_character": 0.000000125, - "output_cost_per_token": 0.0000015, - "output_cost_per_character": 0.000000375, + "input_cost_per_token": 5e-07, + "input_cost_per_character": 1.25e-07, + "output_cost_per_token": 1.5e-06, + "output_cost_per_character": 3.75e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": true, + "supports_parallel_function_calling": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#google_models", "supports_tool_choice": true }, - "gemini-1.0-pro-001": { + "gemini-1.0-pro-001": { "max_tokens": 8192, "max_input_tokens": 32760, "max_output_tokens": 8192, "input_cost_per_image": 0.0025, "input_cost_per_video_per_second": 0.002, - "input_cost_per_token": 0.0000005, - "input_cost_per_character": 0.000000125, - "output_cost_per_token": 0.0000015, - "output_cost_per_character": 0.000000375, + "input_cost_per_token": 5e-07, + "input_cost_per_character": 1.25e-07, + "output_cost_per_token": 1.5e-06, + "output_cost_per_character": 3.75e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", "deprecation_date": "2025-04-09", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-1.0-ultra": { "max_tokens": 8192, @@ -5295,15 +5583,16 @@ "max_output_tokens": 2048, "input_cost_per_image": 0.0025, "input_cost_per_video_per_second": 0.002, - "input_cost_per_token": 0.0000005, - "input_cost_per_character": 0.000000125, - "output_cost_per_token": 0.0000015, - "output_cost_per_character": 0.000000375, + "input_cost_per_token": 5e-07, + "input_cost_per_character": 1.25e-07, + "output_cost_per_token": 1.5e-06, + "output_cost_per_character": 3.75e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": true, "source": "As of Jun, 2024. There is no available doc on vertex ai pricing gemini-1.0-ultra-001. Using gemini-1.0-pro pricing. Got max_tokens info here: https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-1.0-ultra-001": { "max_tokens": 8192, @@ -5311,193 +5600,201 @@ "max_output_tokens": 2048, "input_cost_per_image": 0.0025, "input_cost_per_video_per_second": 0.002, - "input_cost_per_token": 0.0000005, - "input_cost_per_character": 0.000000125, - "output_cost_per_token": 0.0000015, - "output_cost_per_character": 0.000000375, + "input_cost_per_token": 5e-07, + "input_cost_per_character": 1.25e-07, + "output_cost_per_token": 1.5e-06, + "output_cost_per_character": 3.75e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": true, "source": "As of Jun, 2024. There is no available doc on vertex ai pricing gemini-1.0-ultra-001. Using gemini-1.0-pro pricing. Got max_tokens info here: https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, - "gemini-1.0-pro-002": { + "gemini-1.0-pro-002": { "max_tokens": 8192, "max_input_tokens": 32760, "max_output_tokens": 8192, "input_cost_per_image": 0.0025, "input_cost_per_video_per_second": 0.002, - "input_cost_per_token": 0.0000005, - "input_cost_per_character": 0.000000125, - "output_cost_per_token": 0.0000015, - "output_cost_per_character": 0.000000375, + "input_cost_per_token": 5e-07, + "input_cost_per_character": 1.25e-07, + "output_cost_per_token": 1.5e-06, + "output_cost_per_character": 3.75e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", "deprecation_date": "2025-04-09", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, - "gemini-1.5-pro": { + "gemini-1.5-pro": { "max_tokens": 8192, "max_input_tokens": 2097152, "max_output_tokens": 8192, "input_cost_per_image": 0.00032875, - "input_cost_per_audio_per_second": 0.00003125, + "input_cost_per_audio_per_second": 3.125e-05, "input_cost_per_video_per_second": 0.00032875, - "input_cost_per_token": 0.00000125, - "input_cost_per_character": 0.0000003125, - "input_cost_per_image_above_128k_tokens": 0.0006575, - "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, - "input_cost_per_audio_per_second_above_128k_tokens": 0.0000625, - "input_cost_per_token_above_128k_tokens": 0.0000025, - "input_cost_per_character_above_128k_tokens": 0.000000625, - "output_cost_per_token": 0.000005, - "output_cost_per_character": 0.00000125, - "output_cost_per_token_above_128k_tokens": 0.00001, - "output_cost_per_character_above_128k_tokens": 0.0000025, + "input_cost_per_token": 1.25e-06, + "input_cost_per_character": 3.125e-07, + "input_cost_per_image_above_128k_tokens": 0.0006575, + "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, + "input_cost_per_audio_per_second_above_128k_tokens": 6.25e-05, + "input_cost_per_token_above_128k_tokens": 2.5e-06, + "input_cost_per_character_above_128k_tokens": 6.25e-07, + "output_cost_per_token": 5e-06, + "output_cost_per_character": 1.25e-06, + "output_cost_per_token_above_128k_tokens": 1e-05, + "output_cost_per_character_above_128k_tokens": 2.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_vision": true, "supports_pdf_input": true, "supports_system_messages": true, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_response_schema": true, - "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" + "supports_tool_choice": true, + "supports_response_schema": true, + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", + "supports_parallel_function_calling": true }, "gemini-1.5-pro-002": { "max_tokens": 8192, "max_input_tokens": 2097152, "max_output_tokens": 8192, "input_cost_per_image": 0.00032875, - "input_cost_per_audio_per_second": 0.00003125, + "input_cost_per_audio_per_second": 3.125e-05, "input_cost_per_video_per_second": 0.00032875, - "input_cost_per_token": 0.00000125, - "input_cost_per_character": 0.0000003125, - "input_cost_per_image_above_128k_tokens": 0.0006575, - "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, - "input_cost_per_audio_per_second_above_128k_tokens": 0.0000625, - "input_cost_per_token_above_128k_tokens": 0.0000025, - "input_cost_per_character_above_128k_tokens": 0.000000625, - "output_cost_per_token": 0.000005, - "output_cost_per_character": 0.00000125, - "output_cost_per_token_above_128k_tokens": 0.00001, - "output_cost_per_character_above_128k_tokens": 0.0000025, + "input_cost_per_token": 1.25e-06, + "input_cost_per_character": 3.125e-07, + "input_cost_per_image_above_128k_tokens": 0.0006575, + "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, + "input_cost_per_audio_per_second_above_128k_tokens": 6.25e-05, + "input_cost_per_token_above_128k_tokens": 2.5e-06, + "input_cost_per_character_above_128k_tokens": 6.25e-07, + "output_cost_per_token": 5e-06, + "output_cost_per_character": 1.25e-06, + "output_cost_per_token_above_128k_tokens": 1e-05, + "output_cost_per_character_above_128k_tokens": 2.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_vision": true, "supports_system_messages": true, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_response_schema": true, + "supports_tool_choice": true, + "supports_response_schema": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-1.5-pro", - "deprecation_date": "2025-09-24" + "deprecation_date": "2025-09-24", + "supports_parallel_function_calling": true }, - "gemini-1.5-pro-001": { + "gemini-1.5-pro-001": { "max_tokens": 8192, "max_input_tokens": 1000000, "max_output_tokens": 8192, "input_cost_per_image": 0.00032875, - "input_cost_per_audio_per_second": 0.00003125, + "input_cost_per_audio_per_second": 3.125e-05, "input_cost_per_video_per_second": 0.00032875, - "input_cost_per_token": 0.00000125, - "input_cost_per_character": 0.0000003125, - "input_cost_per_image_above_128k_tokens": 0.0006575, - "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, - "input_cost_per_audio_per_second_above_128k_tokens": 0.0000625, - "input_cost_per_token_above_128k_tokens": 0.0000025, - "input_cost_per_character_above_128k_tokens": 0.000000625, - "output_cost_per_token": 0.000005, - "output_cost_per_character": 0.00000125, - "output_cost_per_token_above_128k_tokens": 0.00001, - "output_cost_per_character_above_128k_tokens": 0.0000025, + "input_cost_per_token": 1.25e-06, + "input_cost_per_character": 3.125e-07, + "input_cost_per_image_above_128k_tokens": 0.0006575, + "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, + "input_cost_per_audio_per_second_above_128k_tokens": 6.25e-05, + "input_cost_per_token_above_128k_tokens": 2.5e-06, + "input_cost_per_character_above_128k_tokens": 6.25e-07, + "output_cost_per_token": 5e-06, + "output_cost_per_character": 1.25e-06, + "output_cost_per_token_above_128k_tokens": 1e-05, + "output_cost_per_character_above_128k_tokens": 2.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_vision": true, "supports_system_messages": true, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_response_schema": true, + "supports_tool_choice": true, + "supports_response_schema": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "deprecation_date": "2025-05-24" + "deprecation_date": "2025-05-24", + "supports_parallel_function_calling": true }, - "gemini-1.5-pro-preview-0514": { + "gemini-1.5-pro-preview-0514": { "max_tokens": 8192, "max_input_tokens": 1000000, "max_output_tokens": 8192, "input_cost_per_image": 0.00032875, - "input_cost_per_audio_per_second": 0.00003125, + "input_cost_per_audio_per_second": 3.125e-05, "input_cost_per_video_per_second": 0.00032875, - "input_cost_per_token": 0.000000078125, - "input_cost_per_character": 0.0000003125, - "input_cost_per_image_above_128k_tokens": 0.0006575, - "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, - "input_cost_per_audio_per_second_above_128k_tokens": 0.0000625, - "input_cost_per_token_above_128k_tokens": 0.00000015625, - "input_cost_per_character_above_128k_tokens": 0.000000625, - "output_cost_per_token": 0.0000003125, - "output_cost_per_character": 0.00000125, - "output_cost_per_token_above_128k_tokens": 0.000000625, - "output_cost_per_character_above_128k_tokens": 0.0000025, + "input_cost_per_token": 7.8125e-08, + "input_cost_per_character": 3.125e-07, + "input_cost_per_image_above_128k_tokens": 0.0006575, + "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, + "input_cost_per_audio_per_second_above_128k_tokens": 6.25e-05, + "input_cost_per_token_above_128k_tokens": 1.5625e-07, + "input_cost_per_character_above_128k_tokens": 6.25e-07, + "output_cost_per_token": 3.125e-07, + "output_cost_per_character": 1.25e-06, + "output_cost_per_token_above_128k_tokens": 6.25e-07, + "output_cost_per_character_above_128k_tokens": 2.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_response_schema": true, - "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" + "supports_tool_choice": true, + "supports_response_schema": true, + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", + "supports_parallel_function_calling": true }, - "gemini-1.5-pro-preview-0215": { + "gemini-1.5-pro-preview-0215": { "max_tokens": 8192, "max_input_tokens": 1000000, "max_output_tokens": 8192, "input_cost_per_image": 0.00032875, - "input_cost_per_audio_per_second": 0.00003125, + "input_cost_per_audio_per_second": 3.125e-05, "input_cost_per_video_per_second": 0.00032875, - "input_cost_per_token": 0.000000078125, - "input_cost_per_character": 0.0000003125, - "input_cost_per_image_above_128k_tokens": 0.0006575, - "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, - "input_cost_per_audio_per_second_above_128k_tokens": 0.0000625, - "input_cost_per_token_above_128k_tokens": 0.00000015625, - "input_cost_per_character_above_128k_tokens": 0.000000625, - "output_cost_per_token": 0.0000003125, - "output_cost_per_character": 0.00000125, - "output_cost_per_token_above_128k_tokens": 0.000000625, - "output_cost_per_character_above_128k_tokens": 0.0000025, + "input_cost_per_token": 7.8125e-08, + "input_cost_per_character": 3.125e-07, + "input_cost_per_image_above_128k_tokens": 0.0006575, + "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, + "input_cost_per_audio_per_second_above_128k_tokens": 6.25e-05, + "input_cost_per_token_above_128k_tokens": 1.5625e-07, + "input_cost_per_character_above_128k_tokens": 6.25e-07, + "output_cost_per_token": 3.125e-07, + "output_cost_per_character": 1.25e-06, + "output_cost_per_token_above_128k_tokens": 6.25e-07, + "output_cost_per_character_above_128k_tokens": 2.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, "supports_function_calling": true, - "supports_tool_choice": true, - "supports_response_schema": true, - "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" + "supports_tool_choice": true, + "supports_response_schema": true, + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", + "supports_parallel_function_calling": true }, "gemini-1.5-pro-preview-0409": { "max_tokens": 8192, "max_input_tokens": 1000000, "max_output_tokens": 8192, "input_cost_per_image": 0.00032875, - "input_cost_per_audio_per_second": 0.00003125, + "input_cost_per_audio_per_second": 3.125e-05, "input_cost_per_video_per_second": 0.00032875, - "input_cost_per_token": 0.000000078125, - "input_cost_per_character": 0.0000003125, - "input_cost_per_image_above_128k_tokens": 0.0006575, - "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, - "input_cost_per_audio_per_second_above_128k_tokens": 0.0000625, - "input_cost_per_token_above_128k_tokens": 0.00000015625, - "input_cost_per_character_above_128k_tokens": 0.000000625, - "output_cost_per_token": 0.0000003125, - "output_cost_per_character": 0.00000125, - "output_cost_per_token_above_128k_tokens": 0.000000625, - "output_cost_per_character_above_128k_tokens": 0.0000025, + "input_cost_per_token": 7.8125e-08, + "input_cost_per_character": 3.125e-07, + "input_cost_per_image_above_128k_tokens": 0.0006575, + "input_cost_per_video_per_second_above_128k_tokens": 0.0006575, + "input_cost_per_audio_per_second_above_128k_tokens": 6.25e-05, + "input_cost_per_token_above_128k_tokens": 1.5625e-07, + "input_cost_per_character_above_128k_tokens": 6.25e-07, + "output_cost_per_token": 3.125e-07, + "output_cost_per_character": 1.25e-06, + "output_cost_per_token_above_128k_tokens": 6.25e-07, + "output_cost_per_character_above_128k_tokens": 2.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": true, "supports_tool_choice": true, - "supports_response_schema": true, - "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" + "supports_response_schema": true, + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", + "supports_parallel_function_calling": true }, "gemini-1.5-flash": { "max_tokens": 8192, @@ -5509,20 +5806,20 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_image": 0.00002, - "input_cost_per_video_per_second": 0.00002, - "input_cost_per_audio_per_second": 0.000002, - "input_cost_per_token": 0.000000075, - "input_cost_per_character": 0.00000001875, - "input_cost_per_token_above_128k_tokens": 0.000001, - "input_cost_per_character_above_128k_tokens": 0.00000025, - "input_cost_per_image_above_128k_tokens": 0.00004, - "input_cost_per_video_per_second_above_128k_tokens": 0.00004, - "input_cost_per_audio_per_second_above_128k_tokens": 0.000004, - "output_cost_per_token": 0.0000003, - "output_cost_per_character": 0.000000075, - "output_cost_per_token_above_128k_tokens": 0.0000006, - "output_cost_per_character_above_128k_tokens": 0.00000015, + "input_cost_per_image": 2e-05, + "input_cost_per_video_per_second": 2e-05, + "input_cost_per_audio_per_second": 2e-06, + "input_cost_per_token": 7.5e-08, + "input_cost_per_character": 1.875e-08, + "input_cost_per_token_above_128k_tokens": 1e-06, + "input_cost_per_character_above_128k_tokens": 2.5e-07, + "input_cost_per_image_above_128k_tokens": 4e-05, + "input_cost_per_video_per_second_above_128k_tokens": 4e-05, + "input_cost_per_audio_per_second_above_128k_tokens": 4e-06, + "output_cost_per_token": 3e-07, + "output_cost_per_character": 7.5e-08, + "output_cost_per_token_above_128k_tokens": 6e-07, + "output_cost_per_character_above_128k_tokens": 1.5e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -5530,7 +5827,8 @@ "supports_vision": true, "supports_response_schema": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-1.5-flash-exp-0827": { "max_tokens": 8192, @@ -5542,20 +5840,20 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_image": 0.00002, - "input_cost_per_video_per_second": 0.00002, - "input_cost_per_audio_per_second": 0.000002, - "input_cost_per_token": 0.000000004688, - "input_cost_per_character": 0.00000001875, - "input_cost_per_token_above_128k_tokens": 0.000001, - "input_cost_per_character_above_128k_tokens": 0.00000025, - "input_cost_per_image_above_128k_tokens": 0.00004, - "input_cost_per_video_per_second_above_128k_tokens": 0.00004, - "input_cost_per_audio_per_second_above_128k_tokens": 0.000004, - "output_cost_per_token": 0.0000000046875, - "output_cost_per_character": 0.00000001875, - "output_cost_per_token_above_128k_tokens": 0.000000009375, - "output_cost_per_character_above_128k_tokens": 0.0000000375, + "input_cost_per_image": 2e-05, + "input_cost_per_video_per_second": 2e-05, + "input_cost_per_audio_per_second": 2e-06, + "input_cost_per_token": 4.688e-09, + "input_cost_per_character": 1.875e-08, + "input_cost_per_token_above_128k_tokens": 1e-06, + "input_cost_per_character_above_128k_tokens": 2.5e-07, + "input_cost_per_image_above_128k_tokens": 4e-05, + "input_cost_per_video_per_second_above_128k_tokens": 4e-05, + "input_cost_per_audio_per_second_above_128k_tokens": 4e-06, + "output_cost_per_token": 4.6875e-09, + "output_cost_per_character": 1.875e-08, + "output_cost_per_token_above_128k_tokens": 9.375e-09, + "output_cost_per_character_above_128k_tokens": 3.75e-08, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -5563,7 +5861,8 @@ "supports_vision": true, "supports_response_schema": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-1.5-flash-002": { "max_tokens": 8192, @@ -5575,20 +5874,20 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_image": 0.00002, - "input_cost_per_video_per_second": 0.00002, - "input_cost_per_audio_per_second": 0.000002, - "input_cost_per_token": 0.000000075, - "input_cost_per_character": 0.00000001875, - "input_cost_per_token_above_128k_tokens": 0.000001, - "input_cost_per_character_above_128k_tokens": 0.00000025, - "input_cost_per_image_above_128k_tokens": 0.00004, - "input_cost_per_video_per_second_above_128k_tokens": 0.00004, - "input_cost_per_audio_per_second_above_128k_tokens": 0.000004, - "output_cost_per_token": 0.0000003, - "output_cost_per_character": 0.000000075, - "output_cost_per_token_above_128k_tokens": 0.0000006, - "output_cost_per_character_above_128k_tokens": 0.00000015, + "input_cost_per_image": 2e-05, + "input_cost_per_video_per_second": 2e-05, + "input_cost_per_audio_per_second": 2e-06, + "input_cost_per_token": 7.5e-08, + "input_cost_per_character": 1.875e-08, + "input_cost_per_token_above_128k_tokens": 1e-06, + "input_cost_per_character_above_128k_tokens": 2.5e-07, + "input_cost_per_image_above_128k_tokens": 4e-05, + "input_cost_per_video_per_second_above_128k_tokens": 4e-05, + "input_cost_per_audio_per_second_above_128k_tokens": 4e-06, + "output_cost_per_token": 3e-07, + "output_cost_per_character": 7.5e-08, + "output_cost_per_token_above_128k_tokens": 6e-07, + "output_cost_per_character_above_128k_tokens": 1.5e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -5597,7 +5896,8 @@ "supports_response_schema": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-1.5-flash", "deprecation_date": "2025-09-24", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-1.5-flash-001": { "max_tokens": 8192, @@ -5609,20 +5909,20 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_image": 0.00002, - "input_cost_per_video_per_second": 0.00002, - "input_cost_per_audio_per_second": 0.000002, - "input_cost_per_token": 0.000000075, - "input_cost_per_character": 0.00000001875, - "input_cost_per_token_above_128k_tokens": 0.000001, - "input_cost_per_character_above_128k_tokens": 0.00000025, - "input_cost_per_image_above_128k_tokens": 0.00004, - "input_cost_per_video_per_second_above_128k_tokens": 0.00004, - "input_cost_per_audio_per_second_above_128k_tokens": 0.000004, - "output_cost_per_token": 0.0000003, - "output_cost_per_character": 0.000000075, - "output_cost_per_token_above_128k_tokens": 0.0000006, - "output_cost_per_character_above_128k_tokens": 0.00000015, + "input_cost_per_image": 2e-05, + "input_cost_per_video_per_second": 2e-05, + "input_cost_per_audio_per_second": 2e-06, + "input_cost_per_token": 7.5e-08, + "input_cost_per_character": 1.875e-08, + "input_cost_per_token_above_128k_tokens": 1e-06, + "input_cost_per_character_above_128k_tokens": 2.5e-07, + "input_cost_per_image_above_128k_tokens": 4e-05, + "input_cost_per_video_per_second_above_128k_tokens": 4e-05, + "input_cost_per_audio_per_second_above_128k_tokens": 4e-06, + "output_cost_per_token": 3e-07, + "output_cost_per_character": 7.5e-08, + "output_cost_per_token_above_128k_tokens": 6e-07, + "output_cost_per_character_above_128k_tokens": 1.5e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -5631,7 +5931,8 @@ "supports_response_schema": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", "deprecation_date": "2025-05-24", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-1.5-flash-preview-0514": { "max_tokens": 8192, @@ -5643,27 +5944,28 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_image": 0.00002, - "input_cost_per_video_per_second": 0.00002, - "input_cost_per_audio_per_second": 0.000002, - "input_cost_per_token": 0.000000075, - "input_cost_per_character": 0.00000001875, - "input_cost_per_token_above_128k_tokens": 0.000001, - "input_cost_per_character_above_128k_tokens": 0.00000025, - "input_cost_per_image_above_128k_tokens": 0.00004, - "input_cost_per_video_per_second_above_128k_tokens": 0.00004, - "input_cost_per_audio_per_second_above_128k_tokens": 0.000004, - "output_cost_per_token": 0.0000000046875, - "output_cost_per_character": 0.00000001875, - "output_cost_per_token_above_128k_tokens": 0.000000009375, - "output_cost_per_character_above_128k_tokens": 0.0000000375, + "input_cost_per_image": 2e-05, + "input_cost_per_video_per_second": 2e-05, + "input_cost_per_audio_per_second": 2e-06, + "input_cost_per_token": 7.5e-08, + "input_cost_per_character": 1.875e-08, + "input_cost_per_token_above_128k_tokens": 1e-06, + "input_cost_per_character_above_128k_tokens": 2.5e-07, + "input_cost_per_image_above_128k_tokens": 4e-05, + "input_cost_per_video_per_second_above_128k_tokens": 4e-05, + "input_cost_per_audio_per_second_above_128k_tokens": 4e-06, + "output_cost_per_token": 4.6875e-09, + "output_cost_per_character": 1.875e-08, + "output_cost_per_token_above_128k_tokens": 9.375e-09, + "output_cost_per_character_above_128k_tokens": 3.75e-08, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, "supports_function_calling": true, "supports_vision": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-pro-experimental": { "max_tokens": 8192, @@ -5676,8 +5978,9 @@ "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": false, - "supports_tool_choice": true, - "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/gemini-experimental" + "supports_tool_choice": true, + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/gemini-experimental", + "supports_parallel_function_calling": true }, "gemini-flash-experimental": { "max_tokens": 8192, @@ -5690,8 +5993,9 @@ "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_function_calling": false, - "supports_tool_choice": true, - "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/gemini-experimental" + "supports_tool_choice": true, + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/gemini-experimental", + "supports_parallel_function_calling": true }, "gemini-pro-vision": { "max_tokens": 2048, @@ -5700,15 +6004,16 @@ "max_images_per_prompt": 16, "max_videos_per_prompt": 1, "max_video_length": 2, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "input_cost_per_image": 0.0025, "litellm_provider": "vertex_ai-vision-models", "mode": "chat", "supports_function_calling": true, "supports_vision": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-1.0-pro-vision": { "max_tokens": 2048, @@ -5717,15 +6022,16 @@ "max_images_per_prompt": 16, "max_videos_per_prompt": 1, "max_video_length": 2, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "input_cost_per_image": 0.0025, "litellm_provider": "vertex_ai-vision-models", "mode": "chat", "supports_function_calling": true, "supports_vision": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-1.0-pro-vision-001": { "max_tokens": 2048, @@ -5734,8 +6040,8 @@ "max_images_per_prompt": 16, "max_videos_per_prompt": 1, "max_video_length": 2, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "input_cost_per_image": 0.0025, "litellm_provider": "vertex_ai-vision-models", "mode": "chat", @@ -5743,14 +6049,15 @@ "supports_vision": true, "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", "deprecation_date": "2025-04-09", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "medlm-medium": { "max_tokens": 8192, "max_input_tokens": 32768, "max_output_tokens": 8192, - "input_cost_per_character": 0.0000005, - "output_cost_per_character": 0.000001, + "input_cost_per_character": 5e-07, + "output_cost_per_character": 1e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5760,8 +6067,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_character": 0.000005, - "output_cost_per_character": 0.000015, + "input_cost_per_character": 5e-06, + "output_cost_per_character": 1.5e-05, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", @@ -5777,10 +6084,10 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_200k_tokens": 0.000015, + "input_cost_per_token": 1.25e-06, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_token_above_200k_tokens": 1.5e-05, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -5791,10 +6098,21 @@ "supports_pdf_input": true, "supports_response_schema": true, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], - "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing" + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing", + "supports_parallel_function_calling": true }, "gemini-2.0-pro-exp-02-05": { "max_tokens": 8192, @@ -5806,10 +6124,10 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_200k_tokens": 0.000015, + "input_cost_per_token": 1.25e-06, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_token_above_200k_tokens": 1.5e-05, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -5820,10 +6138,21 @@ "supports_pdf_input": true, "supports_response_schema": true, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], - "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing" + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing", + "supports_parallel_function_calling": true }, "gemini-2.0-flash-exp": { "max_tokens": 8192, @@ -5838,14 +6167,14 @@ "input_cost_per_image": 0, "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, - "input_cost_per_token": 0.00000015, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_token": 1.5e-07, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, - "output_cost_per_token": 0.0000006, + "output_cost_per_token": 6e-07, "output_cost_per_character": 0, "output_cost_per_token_above_128k_tokens": 0, "output_cost_per_character_above_128k_tokens": 0, @@ -5856,10 +6185,19 @@ "supports_vision": true, "supports_response_schema": true, "supports_audio_output": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-2.0-flash-001": { "max_tokens": 8192, @@ -5871,9 +6209,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.000001, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000006, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -5882,10 +6220,19 @@ "supports_response_schema": true, "supports_audio_output": true, "supports_tool_choice": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing", - "deprecation_date": "2026-02-05" + "deprecation_date": "2026-02-05", + "supports_parallel_function_calling": true }, "gemini-2.0-flash-thinking-exp": { "max_tokens": 8192, @@ -5901,9 +6248,9 @@ "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, "input_cost_per_token": 0, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, @@ -5918,10 +6265,19 @@ "supports_vision": true, "supports_response_schema": true, "supports_audio_output": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-2.0-flash", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-2.0-flash-thinking-exp-01-21": { "max_tokens": 65536, @@ -5937,9 +6293,9 @@ "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, "input_cost_per_token": 0, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, @@ -5954,10 +6310,19 @@ "supports_vision": true, "supports_response_schema": false, "supports_audio_output": false, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-2.0-flash", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini/gemini-2.5-pro-exp-03-25": { "max_tokens": 65535, @@ -5985,9 +6350,19 @@ "supports_pdf_input": true, "supports_response_schema": true, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing" }, "gemini/gemini-2.5-flash-preview-tts": { @@ -6000,10 +6375,10 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 1e-6, - "input_cost_per_token": 0.15e-6, - "output_cost_per_token": 0.6e-6, - "output_cost_per_reasoning_token": 3.5e-6, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "output_cost_per_reasoning_token": 3.5e-06, "litellm_provider": "gemini", "mode": "chat", "rpm": 10, @@ -6015,9 +6390,16 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions"], - "supported_modalities": ["text"], - "supported_output_modalities": ["audio"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions" + ], + "supported_modalities": [ + "text" + ], + "supported_output_modalities": [ + "audio" + ], "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview" }, "gemini/gemini-2.5-flash-preview-05-20": { @@ -6030,10 +6412,10 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 1e-6, - "input_cost_per_token": 0.15e-6, - "output_cost_per_token": 0.6e-6, - "output_cost_per_reasoning_token": 3.5e-6, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "output_cost_per_reasoning_token": 3.5e-06, "litellm_provider": "gemini", "mode": "chat", "rpm": 10, @@ -6045,9 +6427,19 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview" }, "gemini/gemini-2.5-flash-preview-04-17": { @@ -6060,10 +6452,10 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 1e-6, - "input_cost_per_token": 0.15e-6, - "output_cost_per_token": 0.6e-6, - "output_cost_per_reasoning_token": 3.5e-6, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "output_cost_per_reasoning_token": 3.5e-06, "litellm_provider": "gemini", "mode": "chat", "rpm": 10, @@ -6075,9 +6467,19 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview" }, "gemini-2.5-flash-preview-05-20": { @@ -6090,10 +6492,10 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 1e-6, - "input_cost_per_token": 0.15e-6, - "output_cost_per_token": 0.6e-6, - "output_cost_per_reasoning_token": 3.5e-6, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "output_cost_per_reasoning_token": 3.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_reasoning": true, @@ -6103,10 +6505,22 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions", "/v1/batch"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], - "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview" + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview", + "supports_parallel_function_calling": true }, "gemini-2.5-flash-preview-04-17": { "max_tokens": 65535, @@ -6118,10 +6532,10 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 1e-6, - "input_cost_per_token": 0.15e-6, - "output_cost_per_token": 0.6e-6, - "output_cost_per_reasoning_token": 3.5e-6, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "output_cost_per_reasoning_token": 3.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_reasoning": true, @@ -6131,10 +6545,22 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions", "/v1/batch"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], - "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview" + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview", + "supports_parallel_function_calling": true }, "gemini-2.0-flash": { "max_tokens": 8192, @@ -6146,9 +6572,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000004, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -6157,10 +6583,19 @@ "supports_response_schema": true, "supports_audio_output": true, "supports_audio_input": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "supports_tool_choice": true, - "source": "https://ai.google.dev/pricing#2_0flash" + "source": "https://ai.google.dev/pricing#2_0flash", + "supports_parallel_function_calling": true }, "gemini-2.0-flash-lite": { "max_input_tokens": 1048576, @@ -6171,9 +6606,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 50, - "input_cost_per_audio_token": 0.000000075, - "input_cost_per_token": 0.000000075, - "output_cost_per_token": 0.0000003, + "input_cost_per_audio_token": 7.5e-08, + "input_cost_per_token": 7.5e-08, + "output_cost_per_token": 3e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -6181,10 +6616,18 @@ "supports_vision": true, "supports_response_schema": true, "supports_audio_output": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-2.0-flash", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_parallel_function_calling": true }, "gemini-2.0-flash-lite-001": { "max_input_tokens": 1048576, @@ -6195,9 +6638,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 50, - "input_cost_per_audio_token": 0.000000075, - "input_cost_per_token": 0.000000075, - "output_cost_per_token": 0.0000003, + "input_cost_per_audio_token": 7.5e-08, + "input_cost_per_token": 7.5e-08, + "output_cost_per_token": 3e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -6205,11 +6648,19 @@ "supports_vision": true, "supports_response_schema": true, "supports_audio_output": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-2.0-flash", "supports_tool_choice": true, - "deprecation_date": "2026-02-25" + "deprecation_date": "2026-02-25", + "supports_parallel_function_calling": true }, "gemini-2.5-pro-preview-05-06": { "max_tokens": 65535, @@ -6221,11 +6672,11 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.00000125, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_200k_tokens": 0.000015, + "input_cost_per_audio_token": 1.25e-06, + "input_cost_per_token": 1.25e-06, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_token_above_200k_tokens": 1.5e-05, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_reasoning": true, @@ -6235,10 +6686,22 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions", "/v1/batch"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], - "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview" + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview", + "supports_parallel_function_calling": true }, "gemini-2.5-pro-preview-03-25": { "max_tokens": 65535, @@ -6250,11 +6713,11 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.00000125, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_200k_tokens": 0.000015, + "input_cost_per_audio_token": 1.25e-06, + "input_cost_per_token": 1.25e-06, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_token_above_200k_tokens": 1.5e-05, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_reasoning": true, @@ -6264,10 +6727,22 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_endpoints": ["/v1/chat/completions", "/v1/completions", "/v1/batch"], - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], - "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview" + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview", + "supports_parallel_function_calling": true }, "gemini-2.0-flash-preview-image-generation": { "max_tokens": 8192, @@ -6279,9 +6754,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000004, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -6290,10 +6765,19 @@ "supports_response_schema": true, "supports_audio_output": true, "supports_audio_input": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "supports_tool_choice": true, - "source": "https://ai.google.dev/pricing#2_0flash" + "source": "https://ai.google.dev/pricing#2_0flash", + "supports_parallel_function_calling": true }, "gemini-2.5-pro-preview-tts": { "max_tokens": 65535, @@ -6305,11 +6789,11 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_200k_tokens": 0.000015, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1.25e-06, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_token_above_200k_tokens": 1.5e-05, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_system_messages": true, @@ -6318,9 +6802,14 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_modalities": ["text"], - "supported_output_modalities": ["audio"], - "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.5-pro-preview" + "supported_modalities": [ + "text" + ], + "supported_output_modalities": [ + "audio" + ], + "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.5-pro-preview", + "supports_parallel_function_calling": true }, "gemini/gemini-2.0-pro-exp-02-05": { "max_tokens": 8192, @@ -6336,9 +6825,9 @@ "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, "input_cost_per_token": 0, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, @@ -6370,9 +6859,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000004, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "gemini", "mode": "chat", "rpm": 10000, @@ -6383,8 +6872,16 @@ "supports_response_schema": true, "supports_audio_output": true, "supports_audio_input": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "supports_tool_choice": true, "source": "https://ai.google.dev/pricing#2_0flash" }, @@ -6398,9 +6895,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000004, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "gemini", "mode": "chat", "rpm": 10000, @@ -6411,8 +6908,16 @@ "supports_response_schema": true, "supports_audio_output": true, "supports_audio_input": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "supports_tool_choice": true, "source": "https://ai.google.dev/pricing#2_0flash" }, @@ -6425,9 +6930,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 50, - "input_cost_per_audio_token": 0.000000075, - "input_cost_per_token": 0.000000075, - "output_cost_per_token": 0.0000003, + "input_cost_per_audio_token": 7.5e-08, + "input_cost_per_token": 7.5e-08, + "output_cost_per_token": 3e-07, "litellm_provider": "gemini", "mode": "chat", "tpm": 4000000, @@ -6438,8 +6943,15 @@ "supports_response_schema": true, "supports_audio_output": true, "supports_tool_choice": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.0-flash-lite" }, "gemini/gemini-2.0-flash-001": { @@ -6452,9 +6964,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000004, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "gemini", "mode": "chat", "rpm": 10000, @@ -6465,8 +6977,16 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "source": "https://ai.google.dev/pricing#2_0flash" }, "gemini/gemini-2.5-pro-preview-tts": { @@ -6479,11 +6999,11 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_200k_tokens": 0.000015, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1.25e-06, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_token_above_200k_tokens": 1.5e-05, "litellm_provider": "gemini", "mode": "chat", "rpm": 10000, @@ -6494,8 +7014,12 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_modalities": ["text"], - "supported_output_modalities": ["audio"], + "supported_modalities": [ + "text" + ], + "supported_output_modalities": [ + "audio" + ], "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.5-pro-preview" }, "gemini/gemini-2.5-pro-preview-05-06": { @@ -6508,11 +7032,11 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_200k_tokens": 0.000015, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1.25e-06, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_token_above_200k_tokens": 1.5e-05, "litellm_provider": "gemini", "mode": "chat", "rpm": 10000, @@ -6523,8 +7047,15 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.5-pro-preview" }, "gemini/gemini-2.5-pro-preview-03-25": { @@ -6537,11 +7068,11 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.00000125, - "input_cost_per_token_above_200k_tokens": 0.0000025, - "output_cost_per_token": 0.00001, - "output_cost_per_token_above_200k_tokens": 0.000015, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1.25e-06, + "input_cost_per_token_above_200k_tokens": 2.5e-06, + "output_cost_per_token": 1e-05, + "output_cost_per_token_above_200k_tokens": 1.5e-05, "litellm_provider": "gemini", "mode": "chat", "rpm": 10000, @@ -6552,8 +7083,15 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.5-pro-preview" }, "gemini/gemini-2.0-flash-exp": { @@ -6570,9 +7108,9 @@ "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, "input_cost_per_token": 0, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, @@ -6589,8 +7127,16 @@ "supports_audio_output": true, "tpm": 4000000, "rpm": 10, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-2.0-flash", "supports_tool_choice": true }, @@ -6604,9 +7150,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.000000075, - "input_cost_per_token": 0.000000075, - "output_cost_per_token": 0.0000003, + "input_cost_per_audio_token": 7.5e-08, + "input_cost_per_token": 7.5e-08, + "output_cost_per_token": 3e-07, "litellm_provider": "gemini", "mode": "chat", "rpm": 60000, @@ -6617,8 +7163,15 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-2.0-flash-lite" }, "gemini/gemini-2.0-flash-thinking-exp": { @@ -6635,9 +7188,9 @@ "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, "input_cost_per_token": 0, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, @@ -6654,8 +7207,16 @@ "supports_audio_output": true, "tpm": 4000000, "rpm": 10, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-2.0-flash", "supports_tool_choice": true }, @@ -6673,9 +7234,9 @@ "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, "input_cost_per_token": 0, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, @@ -6692,8 +7253,16 @@ "supports_audio_output": true, "tpm": 4000000, "rpm": 10, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#gemini-2.0-flash", "supports_tool_choice": true }, @@ -6705,9 +7274,9 @@ "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, "input_cost_per_token": 0, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, @@ -6733,9 +7302,9 @@ "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, "input_cost_per_token": 0, - "input_cost_per_character": 0, - "input_cost_per_token_above_128k_tokens": 0, - "input_cost_per_character_above_128k_tokens": 0, + "input_cost_per_character": 0, + "input_cost_per_token_above_128k_tokens": 0, + "input_cost_per_character_above_128k_tokens": 0, "input_cost_per_image_above_128k_tokens": 0, "input_cost_per_video_per_second_above_128k_tokens": 0, "input_cost_per_audio_per_second_above_128k_tokens": 0, @@ -6757,8 +7326,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6770,8 +7339,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6784,8 +7353,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6798,8 +7367,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6813,8 +7382,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6828,8 +7397,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6843,10 +7412,10 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6864,15 +7433,15 @@ "max_tokens": 32000, "max_input_tokens": 200000, "max_output_tokens": 32000, - "input_cost_per_token": 15e-6, - "output_cost_per_token": 75e-6, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 18.75e-6, - "cache_read_input_token_cost": 1.5e-6, + "cache_creation_input_token_cost": 1.875e-05, + "cache_read_input_token_cost": 1.5e-06, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6890,15 +7459,15 @@ "max_tokens": 64000, "max_input_tokens": 200000, "max_output_tokens": 64000, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 15e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 3.75e-6, - "cache_read_input_token_cost": 0.3e-6, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6913,11 +7482,11 @@ "supports_computer_use": true }, "vertex_ai/claude-3-haiku": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6926,11 +7495,11 @@ "supports_tool_choice": true }, "vertex_ai/claude-3-haiku@20240307": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6942,8 +7511,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000005, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 5e-06, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6955,8 +7524,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000005, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 5e-06, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6968,8 +7537,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6981,8 +7550,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "litellm_provider": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -7002,60 +7571,84 @@ "supports_tool_choice": true }, "vertex_ai/meta/llama-4-scout-17b-16e-instruct-maas": { - "max_tokens": 10e6, - "max_input_tokens": 10e6, - "max_output_tokens": 10e6, - "input_cost_per_token": 0.25e-6, - "output_cost_per_token": 0.70e-6, + "max_tokens": 10000000.0, + "max_input_tokens": 10000000.0, + "max_output_tokens": 10000000.0, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "vertex_ai-llama_models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#partner-models", "supports_tool_choice": true, "supports_function_calling": true, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text", "code"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] }, "vertex_ai/meta/llama-4-scout-17b-128e-instruct-maas": { - "max_tokens": 10e6, - "max_input_tokens": 10e6, - "max_output_tokens": 10e6, - "input_cost_per_token": 0.25e-6, - "output_cost_per_token": 0.70e-6, + "max_tokens": 10000000.0, + "max_input_tokens": 10000000.0, + "max_output_tokens": 10000000.0, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "vertex_ai-llama_models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#partner-models", "supports_tool_choice": true, "supports_function_calling": true, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text", "code"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] }, "vertex_ai/meta/llama-4-maverick-17b-128e-instruct-maas": { - "max_tokens": 1e6, - "max_input_tokens": 1e6, - "max_output_tokens": 1e6, - "input_cost_per_token": 0.35e-6, - "output_cost_per_token": 1.15e-6, + "max_tokens": 1000000.0, + "max_input_tokens": 1000000.0, + "max_output_tokens": 1000000.0, + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 1.15e-06, "litellm_provider": "vertex_ai-llama_models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#partner-models", "supports_tool_choice": true, "supports_function_calling": true, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text", "code"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] }, "vertex_ai/meta/llama-4-maverick-17b-16e-instruct-maas": { - "max_tokens": 1e6, - "max_input_tokens": 1e6, - "max_output_tokens": 1e6, - "input_cost_per_token": 0.35e-6, - "output_cost_per_token": 1.15e-6, + "max_tokens": 1000000.0, + "max_input_tokens": 1000000.0, + "max_output_tokens": 1000000.0, + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 1.15e-06, "litellm_provider": "vertex_ai-llama_models", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#partner-models", "supports_tool_choice": true, "supports_function_calling": true, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text", "code"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] }, "vertex_ai/meta/llama3-70b-instruct-maas": { "max_tokens": 32000, @@ -7096,8 +7689,8 @@ "max_tokens": 8191, "max_input_tokens": 128000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7107,8 +7700,8 @@ "max_tokens": 8191, "max_input_tokens": 128000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7118,8 +7711,8 @@ "max_tokens": 8191, "max_input_tokens": 128000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7129,8 +7722,8 @@ "max_tokens": 8191, "max_input_tokens": 128000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000006, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 6e-06, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7140,8 +7733,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7151,8 +7744,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "vertex_ai-mistral_models", "supports_function_calling": true, "mode": "chat", @@ -7162,8 +7755,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7174,8 +7767,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "vertex_ai-ai21_models", "mode": "chat", "supports_tool_choice": true @@ -7184,8 +7777,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000008, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, "litellm_provider": "vertex_ai-ai21_models", "mode": "chat", "supports_tool_choice": true @@ -7194,8 +7787,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "vertex_ai-ai21_models", "mode": "chat", "supports_tool_choice": true @@ -7204,8 +7797,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "vertex_ai-ai21_models", "mode": "chat", "supports_tool_choice": true @@ -7214,8 +7807,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000008, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, "litellm_provider": "vertex_ai-ai21_models", "mode": "chat", "supports_tool_choice": true @@ -7224,8 +7817,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7235,8 +7828,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7246,8 +7839,8 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, @@ -7257,15 +7850,15 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 128000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "vertex_ai-mistral_models", "mode": "chat", "supports_function_calling": true, "supports_tool_choice": true }, "vertex_ai/imagegeneration@006": { - "output_cost_per_image": 0.020, + "output_cost_per_image": 0.02, "litellm_provider": "vertex_ai-image-models", "mode": "image_generation", "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing" @@ -7292,8 +7885,8 @@ "max_tokens": 2048, "max_input_tokens": 2048, "output_vector_size": 768, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7303,8 +7896,8 @@ "max_tokens": 2048, "max_input_tokens": 2048, "output_vector_size": 768, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7314,8 +7907,8 @@ "max_tokens": 2048, "max_input_tokens": 2048, "output_vector_size": 768, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7325,42 +7918,54 @@ "max_tokens": 2048, "max_input_tokens": 2048, "output_vector_size": 768, - "input_cost_per_character": 0.0000002, + "input_cost_per_character": 2e-07, "input_cost_per_image": 0.0001, "input_cost_per_video_per_second": 0.0005, - "input_cost_per_video_per_second_above_8s_interval": 0.0010, - "input_cost_per_video_per_second_above_15s_interval": 0.0020, - "input_cost_per_token": 0.0000008, + "input_cost_per_video_per_second_above_8s_interval": 0.001, + "input_cost_per_video_per_second_above_15s_interval": 0.002, + "input_cost_per_token": 8e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", - "supported_endpoints": ["/v1/embeddings"], - "supported_modalities": ["text", "image", "video"], + "supported_endpoints": [ + "/v1/embeddings" + ], + "supported_modalities": [ + "text", + "image", + "video" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models" }, "multimodalembedding@001": { "max_tokens": 2048, "max_input_tokens": 2048, "output_vector_size": 768, - "input_cost_per_character": 0.0000002, + "input_cost_per_character": 2e-07, "input_cost_per_image": 0.0001, "input_cost_per_video_per_second": 0.0005, - "input_cost_per_video_per_second_above_8s_interval": 0.0010, - "input_cost_per_video_per_second_above_15s_interval": 0.0020, - "input_cost_per_token": 0.0000008, + "input_cost_per_video_per_second_above_8s_interval": 0.001, + "input_cost_per_video_per_second_above_15s_interval": 0.002, + "input_cost_per_token": 8e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", - "supported_endpoints": ["/v1/embeddings"], - "supported_modalities": ["text", "image", "video"], + "supported_endpoints": [ + "/v1/embeddings" + ], + "supported_modalities": [ + "text", + "image", + "video" + ], "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models" }, "text-embedding-large-exp-03-07": { "max_tokens": 8192, "max_input_tokens": 8192, "output_vector_size": 3072, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7370,8 +7975,8 @@ "max_tokens": 3072, "max_input_tokens": 3072, "output_vector_size": 768, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7381,8 +7986,8 @@ "max_tokens": 3072, "max_input_tokens": 3072, "output_vector_size": 768, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7392,8 +7997,8 @@ "max_tokens": 3072, "max_input_tokens": 3072, "output_vector_size": 768, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7403,8 +8008,8 @@ "max_tokens": 3072, "max_input_tokens": 3072, "output_vector_size": 768, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7414,8 +8019,8 @@ "max_tokens": 3072, "max_input_tokens": 3072, "output_vector_size": 768, - "input_cost_per_character": 0.000000025, - "input_cost_per_token": 0.0000001, + "input_cost_per_character": 2.5e-08, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7425,18 +8030,18 @@ "max_tokens": 3072, "max_input_tokens": 3072, "output_vector_size": 768, - "input_cost_per_token": 0.00000000625, - "input_cost_per_token_batch_requests": 0.000000005, + "input_cost_per_token": 6.25e-09, + "input_cost_per_token_batch_requests": 5e-09, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing" }, - "text-multilingual-embedding-preview-0409":{ + "text-multilingual-embedding-preview-0409": { "max_tokens": 3072, "max_input_tokens": 3072, "output_vector_size": 768, - "input_cost_per_token": 0.00000000625, + "input_cost_per_token": 6.25e-09, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -7446,8 +8051,8 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "palm", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -7456,8 +8061,8 @@ "max_tokens": 4096, "max_input_tokens": 8192, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "palm", "mode": "chat", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -7466,8 +8071,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "palm", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -7476,8 +8081,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "palm", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -7486,8 +8091,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "palm", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -7496,8 +8101,8 @@ "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000125, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 1.25e-07, "litellm_provider": "palm", "mode": "completion", "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models" @@ -7511,13 +8116,13 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, - "cache_read_input_token_cost": 0.00000001875, - "cache_creation_input_token_cost": 0.000001, - "input_cost_per_token": 0.000000075, - "input_cost_per_token_above_128k_tokens": 0.00000015, - "output_cost_per_token": 0.0000003, - "output_cost_per_token_above_128k_tokens": 0.0000006, + "max_pdf_size_mb": 30, + "cache_read_input_token_cost": 1.875e-08, + "cache_creation_input_token_cost": 1e-06, + "input_cost_per_token": 7.5e-08, + "input_cost_per_token_above_128k_tokens": 1.5e-07, + "output_cost_per_token": 3e-07, + "output_cost_per_token_above_128k_tokens": 6e-07, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, @@ -7540,13 +8145,13 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, - "cache_read_input_token_cost": 0.00000001875, - "cache_creation_input_token_cost": 0.000001, - "input_cost_per_token": 0.000000075, - "input_cost_per_token_above_128k_tokens": 0.00000015, - "output_cost_per_token": 0.0000003, - "output_cost_per_token_above_128k_tokens": 0.0000006, + "max_pdf_size_mb": 30, + "cache_read_input_token_cost": 1.875e-08, + "cache_creation_input_token_cost": 1e-06, + "input_cost_per_token": 7.5e-08, + "input_cost_per_token_above_128k_tokens": 1.5e-07, + "output_cost_per_token": 3e-07, + "output_cost_per_token_above_128k_tokens": 6e-07, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, @@ -7569,17 +8174,17 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, - "input_cost_per_token": 0.000000075, - "input_cost_per_token_above_128k_tokens": 0.00000015, - "output_cost_per_token": 0.0000003, - "output_cost_per_token_above_128k_tokens": 0.0000006, + "max_pdf_size_mb": 30, + "input_cost_per_token": 7.5e-08, + "input_cost_per_token_above_128k_tokens": 1.5e-07, + "output_cost_per_token": 3e-07, + "output_cost_per_token_above_128k_tokens": 6e-07, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, "supports_function_calling": true, "supports_vision": true, - "supports_response_schema": true, + "supports_response_schema": true, "tpm": 4000000, "rpm": 2000, "source": "https://ai.google.dev/pricing", @@ -7594,11 +8199,11 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, - "input_cost_per_token": 0.000000075, - "input_cost_per_token_above_128k_tokens": 0.00000015, - "output_cost_per_token": 0.0000003, - "output_cost_per_token_above_128k_tokens": 0.0000006, + "max_pdf_size_mb": 30, + "input_cost_per_token": 7.5e-08, + "input_cost_per_token_above_128k_tokens": 1.5e-07, + "output_cost_per_token": 3e-07, + "output_cost_per_token_above_128k_tokens": 6e-07, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, @@ -7620,7 +8225,7 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, + "max_pdf_size_mb": 30, "input_cost_per_token": 0, "input_cost_per_token_above_128k_tokens": 0, "output_cost_per_token": 0, @@ -7646,7 +8251,7 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, + "max_pdf_size_mb": 30, "input_cost_per_token": 0, "input_cost_per_token_above_128k_tokens": 0, "output_cost_per_token": 0, @@ -7672,7 +8277,7 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, + "max_pdf_size_mb": 30, "input_cost_per_token": 0, "input_cost_per_token_above_128k_tokens": 0, "output_cost_per_token": 0, @@ -7701,7 +8306,7 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, + "max_pdf_size_mb": 30, "input_cost_per_token": 0, "input_cost_per_token_above_128k_tokens": 0, "output_cost_per_token": 0, @@ -7730,7 +8335,7 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, + "max_pdf_size_mb": 30, "input_cost_per_token": 0, "input_cost_per_token_above_128k_tokens": 0, "output_cost_per_token": 0, @@ -7755,7 +8360,7 @@ "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, - "max_pdf_size_mb": 30, + "max_pdf_size_mb": 30, "input_cost_per_token": 0, "input_cost_per_token_above_128k_tokens": 0, "output_cost_per_token": 0, @@ -7775,10 +8380,10 @@ "max_tokens": 8192, "max_input_tokens": 32760, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000035, - "input_cost_per_token_above_128k_tokens": 0.0000007, - "output_cost_per_token": 0.00000105, - "output_cost_per_token_above_128k_tokens": 0.0000021, + "input_cost_per_token": 3.5e-07, + "input_cost_per_token_above_128k_tokens": 7e-07, + "output_cost_per_token": 1.05e-06, + "output_cost_per_token_above_128k_tokens": 2.1e-06, "litellm_provider": "gemini", "mode": "chat", "supports_function_calling": true, @@ -7792,17 +8397,17 @@ "max_tokens": 8192, "max_input_tokens": 2097152, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000035, - "input_cost_per_token_above_128k_tokens": 0.000007, - "output_cost_per_token": 0.0000105, - "output_cost_per_token_above_128k_tokens": 0.000021, + "input_cost_per_token": 3.5e-06, + "input_cost_per_token_above_128k_tokens": 7e-06, + "output_cost_per_token": 1.05e-05, + "output_cost_per_token_above_128k_tokens": 2.1e-05, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, "supports_function_calling": true, "supports_vision": true, - "supports_tool_choice": true, - "supports_response_schema": true, + "supports_tool_choice": true, + "supports_response_schema": true, "tpm": 4000000, "rpm": 1000, "source": "https://ai.google.dev/pricing" @@ -7811,17 +8416,17 @@ "max_tokens": 8192, "max_input_tokens": 2097152, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000035, - "input_cost_per_token_above_128k_tokens": 0.000007, - "output_cost_per_token": 0.0000105, - "output_cost_per_token_above_128k_tokens": 0.000021, + "input_cost_per_token": 3.5e-06, + "input_cost_per_token_above_128k_tokens": 7e-06, + "output_cost_per_token": 1.05e-05, + "output_cost_per_token_above_128k_tokens": 2.1e-05, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, "supports_function_calling": true, "supports_vision": true, - "supports_tool_choice": true, - "supports_response_schema": true, + "supports_tool_choice": true, + "supports_response_schema": true, "supports_prompt_caching": true, "tpm": 4000000, "rpm": 1000, @@ -7832,17 +8437,17 @@ "max_tokens": 8192, "max_input_tokens": 2097152, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000035, - "input_cost_per_token_above_128k_tokens": 0.000007, - "output_cost_per_token": 0.0000105, - "output_cost_per_token_above_128k_tokens": 0.000021, + "input_cost_per_token": 3.5e-06, + "input_cost_per_token_above_128k_tokens": 7e-06, + "output_cost_per_token": 1.05e-05, + "output_cost_per_token_above_128k_tokens": 2.1e-05, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, "supports_function_calling": true, "supports_vision": true, - "supports_tool_choice": true, - "supports_response_schema": true, + "supports_tool_choice": true, + "supports_response_schema": true, "supports_prompt_caching": true, "tpm": 4000000, "rpm": 1000, @@ -7853,10 +8458,10 @@ "max_tokens": 8192, "max_input_tokens": 2097152, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000035, - "input_cost_per_token_above_128k_tokens": 0.000007, - "output_cost_per_token": 0.0000105, - "output_cost_per_token_above_128k_tokens": 0.000021, + "input_cost_per_token": 3.5e-06, + "input_cost_per_token_above_128k_tokens": 7e-06, + "output_cost_per_token": 1.05e-05, + "output_cost_per_token_above_128k_tokens": 2.1e-05, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, @@ -7891,17 +8496,17 @@ "max_tokens": 8192, "max_input_tokens": 1048576, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000035, - "input_cost_per_token_above_128k_tokens": 0.000007, - "output_cost_per_token": 0.00000105, - "output_cost_per_token_above_128k_tokens": 0.000021, + "input_cost_per_token": 3.5e-06, + "input_cost_per_token_above_128k_tokens": 7e-06, + "output_cost_per_token": 1.05e-06, + "output_cost_per_token_above_128k_tokens": 2.1e-05, "litellm_provider": "gemini", "mode": "chat", "supports_system_messages": true, "supports_function_calling": true, "supports_vision": true, - "supports_tool_choice": true, - "supports_response_schema": true, + "supports_tool_choice": true, + "supports_response_schema": true, "tpm": 4000000, "rpm": 1000, "source": "https://ai.google.dev/pricing" @@ -7910,10 +8515,10 @@ "max_tokens": 2048, "max_input_tokens": 30720, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000035, - "input_cost_per_token_above_128k_tokens": 0.0000007, - "output_cost_per_token": 0.00000105, - "output_cost_per_token_above_128k_tokens": 0.0000021, + "input_cost_per_token": 3.5e-07, + "input_cost_per_token_above_128k_tokens": 7e-07, + "output_cost_per_token": 1.05e-06, + "output_cost_per_token_above_128k_tokens": 2.1e-06, "litellm_provider": "gemini", "mode": "chat", "supports_function_calling": true, @@ -7927,8 +8532,8 @@ "gemini/gemini-gemma-2-27b-it": { "max_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000035, - "output_cost_per_token": 0.00000105, + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 1.05e-06, "litellm_provider": "gemini", "mode": "chat", "supports_function_calling": true, @@ -7939,8 +8544,8 @@ "gemini/gemini-gemma-2-9b-it": { "max_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000035, - "output_cost_per_token": 0.00000105, + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 1.05e-06, "litellm_provider": "gemini", "mode": "chat", "supports_function_calling": true, @@ -7952,8 +8557,8 @@ "max_tokens": 8000, "max_input_tokens": 256000, "max_output_tokens": 8000, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "cohere_chat", "mode": "chat", "supports_function_calling": true, @@ -7963,8 +8568,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "cohere_chat", "mode": "chat", "supports_function_calling": true, @@ -7974,8 +8579,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "cohere_chat", "mode": "chat", "supports_function_calling": true, @@ -7985,8 +8590,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000000375, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 3.75e-08, "litellm_provider": "cohere_chat", "mode": "chat", "supports_function_calling": true, @@ -7997,8 +8602,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "cohere_chat", "mode": "chat", "supports_tool_choice": true @@ -8007,8 +8612,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "cohere_chat", "mode": "chat", "supports_function_calling": true, @@ -8018,28 +8623,28 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "cohere_chat", "mode": "chat", "supports_function_calling": true, "supports_tool_choice": true }, "command-nightly": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "cohere", "mode": "completion" }, - "command": { - "max_tokens": 4096, + "command": { + "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "cohere", "mode": "completion" }, @@ -8099,52 +8704,52 @@ "mode": "rerank" }, "embed-english-light-v3.0": { - "max_tokens": 1024, + "max_tokens": 1024, "max_input_tokens": 1024, - "input_cost_per_token": 0.00000010, - "output_cost_per_token": 0.00000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "cohere", "mode": "embedding" }, "embed-multilingual-v3.0": { - "max_tokens": 1024, + "max_tokens": 1024, "max_input_tokens": 1024, - "input_cost_per_token": 0.00000010, - "output_cost_per_token": 0.00000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "cohere", "supports_embedding_image_input": true, "mode": "embedding" }, "embed-english-v2.0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 4096, - "input_cost_per_token": 0.00000010, - "output_cost_per_token": 0.00000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "cohere", "mode": "embedding" }, "embed-english-light-v2.0": { - "max_tokens": 1024, + "max_tokens": 1024, "max_input_tokens": 1024, - "input_cost_per_token": 0.00000010, - "output_cost_per_token": 0.00000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "cohere", "mode": "embedding" }, "embed-multilingual-v2.0": { - "max_tokens": 768, + "max_tokens": 768, "max_input_tokens": 768, - "input_cost_per_token": 0.00000010, - "output_cost_per_token": 0.00000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "cohere", "mode": "embedding" }, "embed-english-v3.0": { - "max_tokens": 1024, + "max_tokens": 1024, "max_input_tokens": 1024, - "input_cost_per_token": 0.00000010, + "input_cost_per_token": 1e-07, "input_cost_per_image": 0.0001, - "output_cost_per_token": 0.00000, + "output_cost_per_token": 0.0, "litellm_provider": "cohere", "mode": "embedding", "supports_image_input": true, @@ -8157,8 +8762,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8167,8 +8772,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8177,8 +8782,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000065, - "output_cost_per_token": 0.00000275, + "input_cost_per_token": 6.5e-07, + "output_cost_per_token": 2.75e-06, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8187,8 +8792,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000065, - "output_cost_per_token": 0.00000275, + "input_cost_per_token": 6.5e-07, + "output_cost_per_token": 2.75e-06, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8197,8 +8802,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000005, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 5e-08, + "output_cost_per_token": 2.5e-07, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8207,8 +8812,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000005, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 5e-08, + "output_cost_per_token": 2.5e-07, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8217,8 +8822,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000065, - "output_cost_per_token": 0.00000275, + "input_cost_per_token": 6.5e-07, + "output_cost_per_token": 2.75e-06, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8227,8 +8832,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000065, - "output_cost_per_token": 0.00000275, + "input_cost_per_token": 6.5e-07, + "output_cost_per_token": 2.75e-06, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8237,8 +8842,8 @@ "max_tokens": 8086, "max_input_tokens": 8086, "max_output_tokens": 8086, - "input_cost_per_token": 0.00000005, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 5e-08, + "output_cost_per_token": 2.5e-07, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8247,8 +8852,8 @@ "max_tokens": 8086, "max_input_tokens": 8086, "max_output_tokens": 8086, - "input_cost_per_token": 0.00000005, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 5e-08, + "output_cost_per_token": 2.5e-07, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8257,8 +8862,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000005, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 5e-08, + "output_cost_per_token": 2.5e-07, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8267,8 +8872,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000005, - "output_cost_per_token": 0.00000025, + "input_cost_per_token": 5e-08, + "output_cost_per_token": 2.5e-07, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8277,8 +8882,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.000001, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 1e-06, "litellm_provider": "replicate", "mode": "chat", "supports_tool_choice": true @@ -8287,12 +8892,12 @@ "max_tokens": 8192, "max_input_tokens": 65336, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000055, - "input_cost_per_token_cache_hit": 0.00000014, - "output_cost_per_token": 0.00000219, + "input_cost_per_token": 5.5e-07, + "input_cost_per_token_cache_hit": 1.4e-07, + "output_cost_per_token": 2.19e-06, "litellm_provider": "openrouter", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_assistant_prefill": true, "supports_reasoning": true, "supports_tool_choice": true, @@ -8302,8 +8907,8 @@ "max_tokens": 8192, "max_input_tokens": 65536, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000014, - "output_cost_per_token": 0.00000028, + "input_cost_per_token": 1.4e-07, + "output_cost_per_token": 2.8e-07, "litellm_provider": "openrouter", "supports_prompt_caching": true, "mode": "chat", @@ -8313,8 +8918,8 @@ "max_tokens": 8192, "max_input_tokens": 66000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000014, - "output_cost_per_token": 0.00000028, + "input_cost_per_token": 1.4e-07, + "output_cost_per_token": 2.8e-07, "litellm_provider": "openrouter", "supports_prompt_caching": true, "mode": "chat", @@ -8322,8 +8927,8 @@ }, "openrouter/microsoft/wizardlm-2-8x22b:nitro": { "max_tokens": 65536, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000001, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true @@ -8332,9 +8937,9 @@ "max_tokens": 8192, "max_input_tokens": 1000000, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.0000075, - "input_cost_per_image": 0.00265, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 7.5e-06, + "input_cost_per_image": 0.00265, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8351,9 +8956,9 @@ "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_audio_token": 0.0000007, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000004, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_system_messages": true, @@ -8365,33 +8970,33 @@ }, "openrouter/mistralai/mixtral-8x22b-instruct": { "max_tokens": 65536, - "input_cost_per_token": 0.00000065, - "output_cost_per_token": 0.00000065, + "input_cost_per_token": 6.5e-07, + "output_cost_per_token": 6.5e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/cohere/command-r-plus": { "max_tokens": 128000, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/databricks/dbrx-instruct": { "max_tokens": 32768, - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/anthropic/claude-3-haiku": { "max_tokens": 200000, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, - "input_cost_per_image": 0.0004, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, + "input_cost_per_image": 0.0004, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8400,8 +9005,8 @@ }, "openrouter/anthropic/claude-3-5-haiku": { "max_tokens": 200000, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000005, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 5e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8411,8 +9016,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8424,8 +9029,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000005, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 5e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8437,8 +9042,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8452,8 +9057,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8466,8 +9071,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "input_cost_per_image": 0.0048, "litellm_provider": "openrouter", "mode": "chat", @@ -8483,8 +9088,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "input_cost_per_image": 0.0048, "litellm_provider": "openrouter", "mode": "chat", @@ -8496,9 +9101,9 @@ }, "openrouter/anthropic/claude-3-sonnet": { "max_tokens": 200000, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "input_cost_per_image": 0.0048, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "input_cost_per_image": 0.0048, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8507,33 +9112,33 @@ }, "openrouter/mistralai/mistral-large": { "max_tokens": 32000, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "mistralai/mistral-small-3.1-24b-instruct": { "max_tokens": 32000, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000003, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 3e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/cognitivecomputations/dolphin-mixtral-8x7b": { "max_tokens": 32769, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/google/gemini-pro-vision": { "max_tokens": 45875, - "input_cost_per_token": 0.000000125, - "output_cost_per_token": 0.000000375, - "input_cost_per_image": 0.0025, + "input_cost_per_token": 1.25e-07, + "output_cost_per_token": 3.75e-07, + "input_cost_per_image": 0.0025, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8542,8 +9147,8 @@ }, "openrouter/fireworks/firellava-13b": { "max_tokens": 4096, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true @@ -8558,24 +9163,24 @@ }, "openrouter/meta-llama/llama-3-8b-instruct:extended": { "max_tokens": 16384, - "input_cost_per_token": 0.000000225, - "output_cost_per_token": 0.00000225, + "input_cost_per_token": 2.25e-07, + "output_cost_per_token": 2.25e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/meta-llama/llama-3-70b-instruct:nitro": { "max_tokens": 8192, - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/meta-llama/llama-3-70b-instruct": { "max_tokens": 8192, - "input_cost_per_token": 0.00000059, - "output_cost_per_token": 0.00000079, + "input_cost_per_token": 5.9e-07, + "output_cost_per_token": 7.9e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true @@ -8584,9 +9189,9 @@ "max_tokens": 100000, "max_input_tokens": 200000, "max_output_tokens": 100000, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.00006, - "cache_read_input_token_cost": 0.0000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, + "cache_read_input_token_cost": 7.5e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8601,8 +9206,8 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000012, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.2e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8614,8 +9219,8 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000012, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.2e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8627,8 +9232,8 @@ "max_tokens": 32768, "max_input_tokens": 128000, "max_output_tokens": 32768, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8640,8 +9245,8 @@ "max_tokens": 32768, "max_input_tokens": 128000, "max_output_tokens": 32768, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000060, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8653,8 +9258,8 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.0000044, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8667,8 +9272,8 @@ "max_tokens": 65536, "max_input_tokens": 128000, "max_output_tokens": 65536, - "input_cost_per_token": 0.0000011, - "output_cost_per_token": 0.0000044, + "input_cost_per_token": 1.1e-06, + "output_cost_per_token": 4.4e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8681,8 +9286,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.000010, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8694,8 +9299,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8705,9 +9310,9 @@ }, "openrouter/openai/gpt-4-vision-preview": { "max_tokens": 130000, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, - "input_cost_per_image": 0.01445, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, + "input_cost_per_image": 0.01445, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8716,24 +9321,24 @@ }, "openrouter/openai/gpt-3.5-turbo": { "max_tokens": 4095, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/openai/gpt-3.5-turbo-16k": { "max_tokens": 16383, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000004, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 4e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/openai/gpt-4": { "max_tokens": 8192, - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.00006, + "input_cost_per_token": 3e-05, + "output_cost_per_token": 6e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true @@ -8741,8 +9346,8 @@ "openrouter/anthropic/claude-instant-v1": { "max_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000163, - "output_cost_per_token": 0.00000551, + "input_cost_per_token": 1.63e-06, + "output_cost_per_token": 5.51e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true @@ -8750,8 +9355,8 @@ "openrouter/anthropic/claude-2": { "max_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00001102, - "output_cost_per_token": 0.00003268, + "input_cost_per_token": 1.102e-05, + "output_cost_per_token": 3.268e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true @@ -8760,8 +9365,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_function_calling": true, @@ -8771,96 +9376,96 @@ }, "openrouter/google/palm-2-chat-bison": { "max_tokens": 25804, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/google/palm-2-codechat-bison": { "max_tokens": 20070, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/meta-llama/llama-2-13b-chat": { "max_tokens": 4096, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/meta-llama/llama-2-70b-chat": { "max_tokens": 4096, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 1.5e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/meta-llama/codellama-34b-instruct": { "max_tokens": 8192, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/nousresearch/nous-hermes-llama2-13b": { "max_tokens": 4096, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/mancer/weaver": { "max_tokens": 8000, - "input_cost_per_token": 0.000005625, - "output_cost_per_token": 0.000005625, + "input_cost_per_token": 5.625e-06, + "output_cost_per_token": 5.625e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/gryphe/mythomax-l2-13b": { "max_tokens": 8192, - "input_cost_per_token": 0.000001875, - "output_cost_per_token": 0.000001875, + "input_cost_per_token": 1.875e-06, + "output_cost_per_token": 1.875e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/jondurbin/airoboros-l2-70b-2.1": { "max_tokens": 4096, - "input_cost_per_token": 0.000013875, - "output_cost_per_token": 0.000013875, + "input_cost_per_token": 1.3875e-05, + "output_cost_per_token": 1.3875e-05, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/undi95/remm-slerp-l2-13b": { "max_tokens": 6144, - "input_cost_per_token": 0.000001875, - "output_cost_per_token": 0.000001875, + "input_cost_per_token": 1.875e-06, + "output_cost_per_token": 1.875e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/pygmalionai/mythalion-13b": { "max_tokens": 4096, - "input_cost_per_token": 0.000001875, - "output_cost_per_token": 0.000001875, + "input_cost_per_token": 1.875e-06, + "output_cost_per_token": 1.875e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true }, "openrouter/mistralai/mistral-7b-instruct": { "max_tokens": 8192, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000013, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 1.3e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true @@ -8877,8 +9482,8 @@ "max_tokens": 33792, "max_input_tokens": 33792, "max_output_tokens": 33792, - "input_cost_per_token": 0.00000018, - "output_cost_per_token": 0.00000018, + "input_cost_per_token": 1.8e-07, + "output_cost_per_token": 1.8e-07, "litellm_provider": "openrouter", "mode": "chat", "supports_tool_choice": true @@ -8887,8 +9492,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 1.5e-05, "litellm_provider": "ai21", "mode": "completion" }, @@ -8896,8 +9501,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "ai21", "mode": "chat", "supports_tool_choice": true @@ -8906,8 +9511,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000008, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, "litellm_provider": "ai21", "mode": "chat", "supports_tool_choice": true @@ -8916,8 +9521,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "ai21", "mode": "chat", "supports_tool_choice": true @@ -8926,8 +9531,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "ai21", "mode": "chat", "supports_tool_choice": true @@ -8936,8 +9541,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000008, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, "litellm_provider": "ai21", "mode": "chat", "supports_tool_choice": true @@ -8946,8 +9551,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000008, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, "litellm_provider": "ai21", "mode": "chat", "supports_tool_choice": true @@ -8956,8 +9561,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "ai21", "mode": "chat", "supports_tool_choice": true @@ -8966,8 +9571,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00001, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 1e-05, "litellm_provider": "ai21", "mode": "completion" }, @@ -8975,8 +9580,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "ai21", "mode": "completion" }, @@ -8984,8 +9589,8 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "nlp_cloud", "mode": "completion" }, @@ -8993,68 +9598,68 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "nlp_cloud", "mode": "chat" }, "luminous-base": { - "max_tokens": 2048, - "input_cost_per_token": 0.00003, - "output_cost_per_token": 0.000033, + "max_tokens": 2048, + "input_cost_per_token": 3e-05, + "output_cost_per_token": 3.3e-05, "litellm_provider": "aleph_alpha", "mode": "completion" }, "luminous-base-control": { - "max_tokens": 2048, - "input_cost_per_token": 0.0000375, - "output_cost_per_token": 0.00004125, + "max_tokens": 2048, + "input_cost_per_token": 3.75e-05, + "output_cost_per_token": 4.125e-05, "litellm_provider": "aleph_alpha", "mode": "chat" }, "luminous-extended": { - "max_tokens": 2048, - "input_cost_per_token": 0.000045, - "output_cost_per_token": 0.0000495, + "max_tokens": 2048, + "input_cost_per_token": 4.5e-05, + "output_cost_per_token": 4.95e-05, "litellm_provider": "aleph_alpha", "mode": "completion" }, "luminous-extended-control": { - "max_tokens": 2048, - "input_cost_per_token": 0.00005625, - "output_cost_per_token": 0.000061875, + "max_tokens": 2048, + "input_cost_per_token": 5.625e-05, + "output_cost_per_token": 6.1875e-05, "litellm_provider": "aleph_alpha", "mode": "chat" }, "luminous-supreme": { - "max_tokens": 2048, + "max_tokens": 2048, "input_cost_per_token": 0.000175, "output_cost_per_token": 0.0001925, "litellm_provider": "aleph_alpha", "mode": "completion" }, "luminous-supreme-control": { - "max_tokens": 2048, + "max_tokens": 2048, "input_cost_per_token": 0.00021875, "output_cost_per_token": 0.000240625, "litellm_provider": "aleph_alpha", "mode": "chat" }, "ai21.j2-mid-v1": { - "max_tokens": 8191, - "max_input_tokens": 8191, - "max_output_tokens": 8191, - "input_cost_per_token": 0.0000125, - "output_cost_per_token": 0.0000125, + "max_tokens": 8191, + "max_input_tokens": 8191, + "max_output_tokens": 8191, + "input_cost_per_token": 1.25e-05, + "output_cost_per_token": 1.25e-05, "litellm_provider": "bedrock", "mode": "chat" }, "ai21.j2-ultra-v1": { - "max_tokens": 8191, - "max_input_tokens": 8191, - "max_output_tokens": 8191, - "input_cost_per_token": 0.0000188, - "output_cost_per_token": 0.0000188, + "max_tokens": 8191, + "max_input_tokens": 8191, + "max_output_tokens": 8191, + "input_cost_per_token": 1.88e-05, + "output_cost_per_token": 1.88e-05, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9062,8 +9667,8 @@ "max_tokens": 4096, "max_input_tokens": 70000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000007, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_system_messages": true @@ -9072,8 +9677,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000008, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9081,8 +9686,8 @@ "max_tokens": 256000, "max_input_tokens": 256000, "max_output_tokens": 256000, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9100,58 +9705,58 @@ "mode": "rerank" }, "amazon.titan-text-lite-v1": { - "max_tokens": 4000, + "max_tokens": 4000, "max_input_tokens": 42000, - "max_output_tokens": 4000, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000004, + "max_output_tokens": 4000, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "bedrock", "mode": "chat" }, "amazon.titan-text-express-v1": { - "max_tokens": 8000, + "max_tokens": 8000, "max_input_tokens": 42000, - "max_output_tokens": 8000, - "input_cost_per_token": 0.0000013, - "output_cost_per_token": 0.0000017, + "max_output_tokens": 8000, + "input_cost_per_token": 1.3e-06, + "output_cost_per_token": 1.7e-06, "litellm_provider": "bedrock", "mode": "chat" }, "amazon.titan-text-premier-v1:0": { - "max_tokens": 32000, + "max_tokens": 32000, "max_input_tokens": 42000, - "max_output_tokens": 32000, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "max_output_tokens": 32000, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "litellm_provider": "bedrock", "mode": "chat" }, "amazon.titan-embed-text-v1": { - "max_tokens": 8192, - "max_input_tokens": 8192, + "max_tokens": 8192, + "max_input_tokens": 8192, "output_vector_size": 1536, - "input_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, "output_cost_per_token": 0.0, - "litellm_provider": "bedrock", + "litellm_provider": "bedrock", "mode": "embedding" }, "amazon.titan-embed-text-v2:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, + "max_tokens": 8192, + "max_input_tokens": 8192, "output_vector_size": 1024, - "input_cost_per_token": 0.0000002, + "input_cost_per_token": 2e-07, "output_cost_per_token": 0.0, - "litellm_provider": "bedrock", + "litellm_provider": "bedrock", "mode": "embedding" }, "amazon.titan-embed-image-v1": { - "max_tokens": 128, - "max_input_tokens": 128, + "max_tokens": 128, + "max_input_tokens": 128, "output_vector_size": 1024, - "input_cost_per_token": 0.0000008, - "input_cost_per_image": 0.00006, + "input_cost_per_token": 8e-07, + "input_cost_per_image": 6e-05, "output_cost_per_token": 0.0, - "litellm_provider": "bedrock", + "litellm_provider": "bedrock", "supports_image_input": true, "supports_embedding_image_input": true, "mode": "embedding", @@ -9164,8 +9769,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9174,8 +9779,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000045, - "output_cost_per_token": 0.0000007, + "input_cost_per_token": 4.5e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9184,8 +9789,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9195,8 +9800,8 @@ "max_tokens": 8191, "max_input_tokens": 128000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000009, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 9e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9206,8 +9811,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000003, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9217,8 +9822,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000045, - "output_cost_per_token": 0.0000007, + "input_cost_per_token": 4.5e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9227,8 +9832,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000045, - "output_cost_per_token": 0.0000007, + "input_cost_per_token": 4.5e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9237,8 +9842,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000059, - "output_cost_per_token": 0.00000091, + "input_cost_per_token": 5.9e-07, + "output_cost_per_token": 9.1e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9247,8 +9852,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9257,8 +9862,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9267,8 +9872,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.00000026, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2.6e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9277,8 +9882,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9288,8 +9893,8 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9299,19 +9904,19 @@ "max_tokens": 8191, "max_input_tokens": 32000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000104, - "output_cost_per_token": 0.0000312, + "input_cost_per_token": 1.04e-05, + "output_cost_per_token": 3.12e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, "supports_tool_choice": true }, "amazon.nova-micro-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 300000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000035, - "output_cost_per_token": 0.00000014, + "input_cost_per_token": 3.5e-08, + "output_cost_per_token": 1.4e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9322,8 +9927,8 @@ "max_tokens": 4096, "max_input_tokens": 300000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000035, - "output_cost_per_token": 0.00000014, + "input_cost_per_token": 3.5e-08, + "output_cost_per_token": 1.4e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9331,11 +9936,11 @@ "supports_response_schema": true }, "eu.amazon.nova-micro-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 300000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000046, - "output_cost_per_token": 0.000000184, + "input_cost_per_token": 4.6e-08, + "output_cost_per_token": 1.84e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9343,11 +9948,11 @@ "supports_response_schema": true }, "amazon.nova-lite-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000006, - "output_cost_per_token": 0.00000024, + "input_cost_per_token": 6e-08, + "output_cost_per_token": 2.4e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9360,8 +9965,8 @@ "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000006, - "output_cost_per_token": 0.00000024, + "input_cost_per_token": 6e-08, + "output_cost_per_token": 2.4e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9371,11 +9976,11 @@ "supports_response_schema": true }, "eu.amazon.nova-lite-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000000078, - "output_cost_per_token": 0.000000312, + "input_cost_per_token": 7.8e-08, + "output_cost_per_token": 3.12e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9385,11 +9990,11 @@ "supports_response_schema": true }, "amazon.nova-pro-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 300000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.0000032, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 3.2e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9402,8 +10007,8 @@ "max_tokens": 4096, "max_input_tokens": 300000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.0000032, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 3.2e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9413,17 +10018,17 @@ "supports_response_schema": true }, "1024-x-1024/50-steps/bedrock/amazon.nova-canvas-v1:0": { - "max_input_tokens": 2600, - "output_cost_per_image": 0.06, - "litellm_provider": "bedrock", - "mode": "image_generation" + "max_input_tokens": 2600, + "output_cost_per_image": 0.06, + "litellm_provider": "bedrock", + "mode": "image_generation" }, "eu.amazon.nova-pro-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 300000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000105, - "output_cost_per_token": 0.0000042, + "input_cost_per_token": 1.05e-06, + "output_cost_per_token": 4.2e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9437,8 +10042,8 @@ "max_tokens": 4096, "max_input_tokens": 1000000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000025, - "output_cost_per_token": 0.0000125, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1.25e-05, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9448,11 +10053,11 @@ "supports_response_schema": true }, "anthropic.claude-3-sonnet-20240229-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9462,11 +10067,11 @@ "supports_tool_choice": true }, "bedrock/invoke/anthropic.claude-3-5-sonnet-20240620-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9478,11 +10083,11 @@ } }, "anthropic.claude-3-5-sonnet-20240620-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9495,15 +10100,15 @@ "max_tokens": 32000, "max_input_tokens": 200000, "max_output_tokens": 32000, - "input_cost_per_token": 15e-6, - "output_cost_per_token": 75e-6, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 18.75e-6, - "cache_read_input_token_cost": 1.5e-6, + "cache_creation_input_token_cost": 1.875e-05, + "cache_read_input_token_cost": 1.5e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9521,15 +10126,15 @@ "max_tokens": 64000, "max_input_tokens": 200000, "max_output_tokens": 64000, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 15e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 3.75e-6, - "cache_read_input_token_cost": 0.3e-6, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9548,16 +10153,16 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, "supports_vision": true, "supports_assistant_prefill": true, - "supports_prompt_caching": true, + "supports_prompt_caching": true, "supports_response_schema": true, "supports_pdf_input": true, "supports_reasoning": true, @@ -9568,26 +10173,26 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, "supports_vision": true, "supports_pdf_input": true, "supports_assistant_prefill": true, - "supports_prompt_caching": true, + "supports_prompt_caching": true, "supports_response_schema": true, "supports_tool_choice": true }, "anthropic.claude-3-haiku-20240307-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9600,10 +10205,10 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.000004, - "cache_creation_input_token_cost": 0.000001, - "cache_read_input_token_cost": 0.00000008, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 4e-06, + "cache_creation_input_token_cost": 1e-06, + "cache_read_input_token_cost": 8e-08, "litellm_provider": "bedrock", "mode": "chat", "supports_assistant_prefill": true, @@ -9617,8 +10222,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9630,8 +10235,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9644,8 +10249,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9659,10 +10264,10 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9678,16 +10283,16 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "cache_creation_input_token_cost": 0.00000375, - "cache_read_input_token_cost": 0.0000003, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, "supports_vision": true, "supports_assistant_prefill": true, - "supports_prompt_caching": true, + "supports_prompt_caching": true, "supports_response_schema": true, "supports_pdf_input": true, "supports_tool_choice": true, @@ -9697,15 +10302,15 @@ "max_tokens": 32000, "max_input_tokens": 200000, "max_output_tokens": 32000, - "input_cost_per_token": 15e-6, - "output_cost_per_token": 75e-6, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 18.75e-6, - "cache_read_input_token_cost": 1.5e-6, + "cache_creation_input_token_cost": 1.875e-05, + "cache_read_input_token_cost": 1.5e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9723,15 +10328,15 @@ "max_tokens": 64000, "max_input_tokens": 200000, "max_output_tokens": 64000, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 15e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 3.75e-6, - "cache_read_input_token_cost": 0.3e-6, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9749,8 +10354,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9763,10 +10368,10 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.000004, - "cache_creation_input_token_cost": 0.000001, - "cache_read_input_token_cost": 0.00000008, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 4e-06, + "cache_creation_input_token_cost": 1e-06, + "cache_read_input_token_cost": 8e-08, "litellm_provider": "bedrock", "mode": "chat", "supports_assistant_prefill": true, @@ -9780,8 +10385,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9793,8 +10398,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9807,8 +10412,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9822,8 +10427,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9839,14 +10444,14 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, "supports_vision": true, "supports_assistant_prefill": true, - "supports_prompt_caching": true, + "supports_prompt_caching": true, "supports_response_schema": true, "supports_pdf_input": true, "supports_tool_choice": true, @@ -9856,8 +10461,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9870,15 +10475,15 @@ "max_tokens": 32000, "max_input_tokens": 200000, "max_output_tokens": 32000, - "input_cost_per_token": 15e-6, - "output_cost_per_token": 75e-6, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 18.75e-6, - "cache_read_input_token_cost": 1.5e-6, + "cache_creation_input_token_cost": 1.875e-05, + "cache_read_input_token_cost": 1.5e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9896,15 +10501,15 @@ "max_tokens": 64000, "max_input_tokens": 200000, "max_output_tokens": 64000, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 15e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "search_context_cost_per_query": { - "search_context_size_low": 1e-2, - "search_context_size_medium": 1e-2, - "search_context_size_high": 1e-2 + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 }, - "cache_creation_input_token_cost": 3.75e-6, - "cache_read_input_token_cost": 0.3e-6, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -9922,8 +10527,8 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000125, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9937,8 +10542,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000015, - "output_cost_per_token": 0.000075, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -9947,46 +10552,46 @@ "supports_tool_choice": true }, "anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-east-1/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/us-west-2/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/ap-northeast-1/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/ap-northeast-1/1-month-commitment/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.0455, @@ -9995,7 +10600,7 @@ "mode": "chat" }, "bedrock/ap-northeast-1/6-month-commitment/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.02527, @@ -10004,112 +10609,112 @@ "mode": "chat" }, "bedrock/eu-central-1/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "max_output_tokens": 8191, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/eu-central-1/1-month-commitment/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.0415, "output_cost_per_second": 0.0415, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/eu-central-1/6-month-commitment/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.02305, "output_cost_per_second": 0.02305, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-east-1/1-month-commitment/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.0175, "output_cost_per_second": 0.0175, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-east-1/6-month-commitment/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.00972, "output_cost_per_second": 0.00972, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-west-2/1-month-commitment/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.0175, "output_cost_per_second": 0.0175, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-west-2/6-month-commitment/anthropic.claude-v1": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.00972, "output_cost_per_second": 0.00972, "litellm_provider": "bedrock", "mode": "chat" }, "anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "max_output_tokens": 8191, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/us-east-1/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "max_output_tokens": 8191, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/us-west-2/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "max_output_tokens": 8191, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/ap-northeast-1/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "max_output_tokens": 8191, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/ap-northeast-1/1-month-commitment/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.0455, "output_cost_per_second": 0.0455, "litellm_provider": "bedrock", @@ -10117,9 +10722,9 @@ "supports_tool_choice": true }, "bedrock/ap-northeast-1/6-month-commitment/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.02527, "output_cost_per_second": 0.02527, "litellm_provider": "bedrock", @@ -10127,19 +10732,19 @@ "supports_tool_choice": true }, "bedrock/eu-central-1/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "max_output_tokens": 8191, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/eu-central-1/1-month-commitment/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.0415, "output_cost_per_second": 0.0415, "litellm_provider": "bedrock", @@ -10147,9 +10752,9 @@ "supports_tool_choice": true }, "bedrock/eu-central-1/6-month-commitment/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.02305, "output_cost_per_second": 0.02305, "litellm_provider": "bedrock", @@ -10157,9 +10762,9 @@ "supports_tool_choice": true }, "bedrock/us-east-1/1-month-commitment/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.0175, "output_cost_per_second": 0.0175, "litellm_provider": "bedrock", @@ -10167,9 +10772,9 @@ "supports_tool_choice": true }, "bedrock/us-east-1/6-month-commitment/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.00972, "output_cost_per_second": 0.00972, "litellm_provider": "bedrock", @@ -10177,9 +10782,9 @@ "supports_tool_choice": true }, "bedrock/us-west-2/1-month-commitment/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.0175, "output_cost_per_second": 0.0175, "litellm_provider": "bedrock", @@ -10187,9 +10792,9 @@ "supports_tool_choice": true }, "bedrock/us-west-2/6-month-commitment/anthropic.claude-v2": { - "max_tokens": 8191, + "max_tokens": 8191, "max_input_tokens": 100000, - "max_output_tokens": 8191, + "max_output_tokens": 8191, "input_cost_per_second": 0.00972, "output_cost_per_second": 0.00972, "litellm_provider": "bedrock", @@ -10197,48 +10802,48 @@ "supports_tool_choice": true }, "anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/us-east-1/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/us-west-2/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/ap-northeast-1/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/ap-northeast-1/1-month-commitment/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.0455, "output_cost_per_second": 0.0455, @@ -10247,8 +10852,8 @@ "supports_tool_choice": true }, "bedrock/ap-northeast-1/6-month-commitment/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.02527, "output_cost_per_second": 0.02527, @@ -10257,18 +10862,18 @@ "supports_tool_choice": true }, "bedrock/eu-central-1/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.000008, - "output_cost_per_token": 0.000024, + "input_cost_per_token": 8e-06, + "output_cost_per_token": 2.4e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/eu-central-1/1-month-commitment/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.0415, "output_cost_per_second": 0.0415, @@ -10277,8 +10882,8 @@ "supports_tool_choice": true }, "bedrock/eu-central-1/6-month-commitment/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.02305, "output_cost_per_second": 0.02305, @@ -10287,8 +10892,8 @@ "supports_tool_choice": true }, "bedrock/us-east-1/1-month-commitment/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.0175, "output_cost_per_second": 0.0175, @@ -10297,8 +10902,8 @@ "supports_tool_choice": true }, "bedrock/us-east-1/6-month-commitment/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.00972, "output_cost_per_second": 0.00972, @@ -10307,8 +10912,8 @@ "supports_tool_choice": true }, "bedrock/us-west-2/1-month-commitment/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.0175, "output_cost_per_second": 0.0175, @@ -10317,8 +10922,8 @@ "supports_tool_choice": true }, "bedrock/us-west-2/6-month-commitment/anthropic.claude-v2:1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.00972, "output_cost_per_second": 0.00972, @@ -10327,28 +10932,28 @@ "supports_tool_choice": true }, "anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.0000024, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 2.4e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/us-east-1/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.0000024, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 2.4e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/us-east-1/1-month-commitment/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.011, "output_cost_per_second": 0.011, @@ -10357,8 +10962,8 @@ "supports_tool_choice": true }, "bedrock/us-east-1/6-month-commitment/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.00611, "output_cost_per_second": 0.00611, @@ -10367,8 +10972,8 @@ "supports_tool_choice": true }, "bedrock/us-west-2/1-month-commitment/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.011, "output_cost_per_second": 0.011, @@ -10377,8 +10982,8 @@ "supports_tool_choice": true }, "bedrock/us-west-2/6-month-commitment/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.00611, "output_cost_per_second": 0.00611, @@ -10387,28 +10992,28 @@ "supports_tool_choice": true }, "bedrock/us-west-2/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.0000024, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 2.4e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/ap-northeast-1/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000223, - "output_cost_per_token": 0.00000755, + "input_cost_per_token": 2.23e-06, + "output_cost_per_token": 7.55e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/ap-northeast-1/1-month-commitment/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.01475, "output_cost_per_second": 0.01475, @@ -10417,8 +11022,8 @@ "supports_tool_choice": true }, "bedrock/ap-northeast-1/6-month-commitment/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.008194, "output_cost_per_second": 0.008194, @@ -10427,18 +11032,18 @@ "supports_tool_choice": true }, "bedrock/eu-central-1/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000248, - "output_cost_per_token": 0.00000838, + "input_cost_per_token": 2.48e-06, + "output_cost_per_token": 8.38e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/eu-central-1/1-month-commitment/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.01635, "output_cost_per_second": 0.01635, @@ -10447,8 +11052,8 @@ "supports_tool_choice": true }, "bedrock/eu-central-1/6-month-commitment/anthropic.claude-instant-v1": { - "max_tokens": 8191, - "max_input_tokens": 100000, + "max_tokens": 8191, + "max_input_tokens": 100000, "max_output_tokens": 8191, "input_cost_per_second": 0.009083, "output_cost_per_second": 0.009083, @@ -10470,19 +11075,19 @@ "mode": "rerank" }, "cohere.command-text-v14": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.0000020, + "max_output_tokens": 4096, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/*/1-month-commitment/cohere.command-text-v14": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_second": 0.011, "output_cost_per_second": 0.011, "litellm_provider": "bedrock", @@ -10490,9 +11095,9 @@ "supports_tool_choice": true }, "bedrock/*/6-month-commitment/cohere.command-text-v14": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_second": 0.0066027, "output_cost_per_second": 0.0066027, "litellm_provider": "bedrock", @@ -10500,19 +11105,19 @@ "supports_tool_choice": true }, "cohere.command-light-text-v14": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000006, + "max_output_tokens": 4096, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "bedrock/*/1-month-commitment/cohere.command-light-text-v14": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_second": 0.001902, "output_cost_per_second": 0.001902, "litellm_provider": "bedrock", @@ -10520,9 +11125,9 @@ "supports_tool_choice": true }, "bedrock/*/6-month-commitment/cohere.command-light-text-v14": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_second": 0.0011416, "output_cost_per_second": 0.0011416, "litellm_provider": "bedrock", @@ -10530,226 +11135,225 @@ "supports_tool_choice": true }, "cohere.command-r-plus-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000030, - "output_cost_per_token": 0.000015, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "cohere.command-r-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000015, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true }, "cohere.embed-english-v3": { - "max_tokens": 512, - "max_input_tokens": 512, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, - "litellm_provider": "bedrock", + "max_tokens": 512, + "max_input_tokens": 512, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, + "litellm_provider": "bedrock", "mode": "embedding", "supports_embedding_image_input": true }, "cohere.embed-multilingual-v3": { - "max_tokens": 512, - "max_input_tokens": 512, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, + "max_tokens": 512, + "max_input_tokens": 512, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "bedrock", "mode": "embedding", "supports_embedding_image_input": true }, "us.deepseek.r1-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000135, - "output_cost_per_token": 0.0000054, + "input_cost_per_token": 1.35e-06, + "output_cost_per_token": 5.4e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_reasoning": true, - "supports_function_calling": false, + "supports_function_calling": false, "supports_tool_choice": false - }, "meta.llama3-3-70b-instruct-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000072, - "output_cost_per_token": 0.00000072, + "input_cost_per_token": 7.2e-07, + "output_cost_per_token": 7.2e-07, "litellm_provider": "bedrock_converse", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "meta.llama2-13b-chat-v1": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000075, - "output_cost_per_token": 0.000001, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 7.5e-07, + "output_cost_per_token": 1e-06, "litellm_provider": "bedrock", "mode": "chat" }, "meta.llama2-70b-chat-v1": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000195, - "output_cost_per_token": 0.00000256, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 1.95e-06, + "output_cost_per_token": 2.56e-06, "litellm_provider": "bedrock", "mode": "chat" }, "meta.llama3-8b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000006, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-east-1/meta.llama3-8b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000006, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-west-1/meta.llama3-8b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000006, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/ap-south-1/meta.llama3-8b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000036, - "output_cost_per_token": 0.00000072, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3.6e-07, + "output_cost_per_token": 7.2e-07, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/ca-central-1/meta.llama3-8b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000035, - "output_cost_per_token": 0.00000069, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 6.9e-07, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/eu-west-1/meta.llama3-8b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000032, - "output_cost_per_token": 0.00000065, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3.2e-07, + "output_cost_per_token": 6.5e-07, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/eu-west-2/meta.llama3-8b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000039, - "output_cost_per_token": 0.00000078, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3.9e-07, + "output_cost_per_token": 7.8e-07, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/sa-east-1/meta.llama3-8b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.00000101, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.01e-06, "litellm_provider": "bedrock", "mode": "chat" }, "meta.llama3-70b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000265, - "output_cost_per_token": 0.0000035, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 2.65e-06, + "output_cost_per_token": 3.5e-06, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-east-1/meta.llama3-70b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000265, - "output_cost_per_token": 0.0000035, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 2.65e-06, + "output_cost_per_token": 3.5e-06, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/us-west-1/meta.llama3-70b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000265, - "output_cost_per_token": 0.0000035, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 2.65e-06, + "output_cost_per_token": 3.5e-06, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/ap-south-1/meta.llama3-70b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000318, - "output_cost_per_token": 0.0000042, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3.18e-06, + "output_cost_per_token": 4.2e-06, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/ca-central-1/meta.llama3-70b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000305, - "output_cost_per_token": 0.00000403, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3.05e-06, + "output_cost_per_token": 4.03e-06, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/eu-west-1/meta.llama3-70b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000286, - "output_cost_per_token": 0.00000378, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 2.86e-06, + "output_cost_per_token": 3.78e-06, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/eu-west-2/meta.llama3-70b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000345, - "output_cost_per_token": 0.00000455, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 3.45e-06, + "output_cost_per_token": 4.55e-06, "litellm_provider": "bedrock", "mode": "chat" }, "bedrock/sa-east-1/meta.llama3-70b-instruct-v1:0": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000445, - "output_cost_per_token": 0.00000588, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 4.45e-06, + "output_cost_per_token": 5.88e-06, "litellm_provider": "bedrock", "mode": "chat" }, @@ -10757,143 +11361,143 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000022, - "output_cost_per_token": 0.00000022, + "input_cost_per_token": 2.2e-07, + "output_cost_per_token": 2.2e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "us.meta.llama3-1-8b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000022, - "output_cost_per_token": 0.00000022, + "input_cost_per_token": 2.2e-07, + "output_cost_per_token": 2.2e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "meta.llama3-1-70b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000099, - "output_cost_per_token": 0.00000099, + "input_cost_per_token": 9.9e-07, + "output_cost_per_token": 9.9e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "us.meta.llama3-1-70b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 2048, - "input_cost_per_token": 0.00000099, - "output_cost_per_token": 0.00000099, + "input_cost_per_token": 9.9e-07, + "output_cost_per_token": 9.9e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "meta.llama3-1-405b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000532, - "output_cost_per_token": 0.000016, + "input_cost_per_token": 5.32e-06, + "output_cost_per_token": 1.6e-05, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "us.meta.llama3-1-405b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000532, - "output_cost_per_token": 0.000016, + "input_cost_per_token": 5.32e-06, + "output_cost_per_token": 1.6e-05, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "meta.llama3-2-1b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 1e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "us.meta.llama3-2-1b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 1e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "eu.meta.llama3-2-1b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000013, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 1.3e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "meta.llama3-2-3b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "us.meta.llama3-2-3b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "eu.meta.llama3-2-3b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000019, - "output_cost_per_token": 0.00000019, + "input_cost_per_token": 1.9e-07, + "output_cost_per_token": 1.9e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "meta.llama3-2-11b-instruct-v1:0": { "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000035, - "output_cost_per_token": 0.00000035, + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 3.5e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false, "supports_vision": true }, @@ -10901,11 +11505,11 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000035, - "output_cost_per_token": 0.00000035, + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 3.5e-07, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false, "supports_vision": true }, @@ -10913,11 +11517,11 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false, "supports_vision": true }, @@ -10925,130 +11529,154 @@ "max_tokens": 128000, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000002, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "bedrock", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false, "supports_vision": true }, "us.meta.llama3-3-70b-instruct-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000072, - "output_cost_per_token": 0.00000072, + "input_cost_per_token": 7.2e-07, + "output_cost_per_token": 7.2e-07, "litellm_provider": "bedrock_converse", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false }, "meta.llama4-maverick-17b-instruct-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00024e-3, - "input_cost_per_token_batches": 0.00012e-3, - "output_cost_per_token": 0.00097e-3, - "output_cost_per_token_batches": 0.000485e-3, + "input_cost_per_token": 2.4e-07, + "input_cost_per_token_batches": 1.2e-07, + "output_cost_per_token": 9.7e-07, + "output_cost_per_token_batches": 4.85e-07, "litellm_provider": "bedrock_converse", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text", "code"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] }, "us.meta.llama4-maverick-17b-instruct-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00024e-3, - "input_cost_per_token_batches": 0.00012e-3, - "output_cost_per_token": 0.00097e-3, - "output_cost_per_token_batches": 0.000485e-3, + "input_cost_per_token": 2.4e-07, + "input_cost_per_token_batches": 1.2e-07, + "output_cost_per_token": 9.7e-07, + "output_cost_per_token_batches": 4.85e-07, "litellm_provider": "bedrock_converse", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text", "code"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] }, "meta.llama4-scout-17b-instruct-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00017e-3, - "input_cost_per_token_batches": 0.000085e-3, - "output_cost_per_token": 0.00066e-3, - "output_cost_per_token_batches": 0.00033e-3, + "input_cost_per_token": 1.7e-07, + "input_cost_per_token_batches": 8.5e-08, + "output_cost_per_token": 6.6e-07, + "output_cost_per_token_batches": 3.3e-07, "litellm_provider": "bedrock_converse", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text", "code"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] }, "us.meta.llama4-scout-17b-instruct-v1:0": { - "max_tokens": 4096, + "max_tokens": 4096, "max_input_tokens": 128000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00017e-3, - "input_cost_per_token_batches": 0.000085e-3, - "output_cost_per_token": 0.00066e-3, - "output_cost_per_token_batches": 0.00033e-3, + "input_cost_per_token": 1.7e-07, + "input_cost_per_token_batches": 8.5e-08, + "output_cost_per_token": 6.6e-07, + "output_cost_per_token_batches": 3.3e-07, "litellm_provider": "bedrock_converse", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": true, "supports_tool_choice": false, - "supported_modalities": ["text", "image"], - "supported_output_modalities": ["text", "code"] + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] }, "512-x-512/50-steps/stability.stable-diffusion-xl-v0": { - "max_tokens": 77, - "max_input_tokens": 77, + "max_tokens": 77, + "max_input_tokens": 77, "output_cost_per_image": 0.018, "litellm_provider": "bedrock", "mode": "image_generation" }, "512-x-512/max-steps/stability.stable-diffusion-xl-v0": { - "max_tokens": 77, - "max_input_tokens": 77, + "max_tokens": 77, + "max_input_tokens": 77, "output_cost_per_image": 0.036, "litellm_provider": "bedrock", "mode": "image_generation" }, "max-x-max/50-steps/stability.stable-diffusion-xl-v0": { - "max_tokens": 77, - "max_input_tokens": 77, + "max_tokens": 77, + "max_input_tokens": 77, "output_cost_per_image": 0.036, "litellm_provider": "bedrock", "mode": "image_generation" }, "max-x-max/max-steps/stability.stable-diffusion-xl-v0": { - "max_tokens": 77, - "max_input_tokens": 77, + "max_tokens": 77, + "max_input_tokens": 77, "output_cost_per_image": 0.072, "litellm_provider": "bedrock", "mode": "image_generation" }, "1024-x-1024/50-steps/stability.stable-diffusion-xl-v1": { - "max_tokens": 77, - "max_input_tokens": 77, + "max_tokens": 77, + "max_input_tokens": 77, "output_cost_per_image": 0.04, "litellm_provider": "bedrock", "mode": "image_generation" }, "1024-x-1024/max-steps/stability.stable-diffusion-xl-v1": { - "max_tokens": 77, - "max_input_tokens": 77, + "max_tokens": 77, + "max_input_tokens": 77, "output_cost_per_image": 0.08, "litellm_provider": "bedrock", "mode": "image_generation" }, "stability.sd3-large-v1:0": { - "max_tokens": 77, - "max_input_tokens": 77, + "max_tokens": 77, + "max_input_tokens": 77, "output_cost_per_image": 0.08, "litellm_provider": "bedrock", "mode": "image_generation" @@ -11075,8 +11703,8 @@ "mode": "image_generation" }, "stability.stable-image-ultra-v1:0": { - "max_tokens": 77, - "max_input_tokens": 77, + "max_tokens": 77, + "max_input_tokens": 77, "output_cost_per_image": 0.14, "litellm_provider": "bedrock", "mode": "image_generation" @@ -11089,111 +11717,111 @@ "mode": "image_generation" }, "sagemaker/meta-textgeneration-llama-2-7b": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "completion" }, "sagemaker/meta-textgeneration-llama-2-7b-f": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "chat" }, "sagemaker/meta-textgeneration-llama-2-13b": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "completion" }, "sagemaker/meta-textgeneration-llama-2-13b-f": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "chat" }, "sagemaker/meta-textgeneration-llama-2-70b": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "completion" }, "sagemaker/meta-textgeneration-llama-2-70b-b-f": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "chat" }, "together-ai-up-to-4b": { - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000001, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 1e-07, "litellm_provider": "together_ai", "mode": "chat" }, "together-ai-4.1b-8b": { - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "together_ai", "mode": "chat" }, "together-ai-8.1b-21b": { "max_tokens": 1000, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000003, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 3e-07, "litellm_provider": "together_ai", "mode": "chat" }, "together-ai-21.1b-41b": { - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.0000008, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 8e-07, "litellm_provider": "together_ai", "mode": "chat" }, "together-ai-41.1b-80b": { - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "together_ai", "mode": "chat" }, "together-ai-81.1b-110b": { - "input_cost_per_token": 0.0000018, - "output_cost_per_token": 0.0000018, + "input_cost_per_token": 1.8e-06, + "output_cost_per_token": 1.8e-06, "litellm_provider": "together_ai", "mode": "chat" }, "together-ai-embedding-up-to-150m": { - "input_cost_per_token": 0.000000008, + "input_cost_per_token": 8e-09, "output_cost_per_token": 0.0, "litellm_provider": "together_ai", "mode": "embedding" }, "together-ai-embedding-151m-to-350m": { - "input_cost_per_token": 0.000000016, + "input_cost_per_token": 1.6e-08, "output_cost_per_token": 0.0, "litellm_provider": "together_ai", "mode": "embedding" }, "together_ai/meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo": { - "input_cost_per_token": 0.00000018, - "output_cost_per_token": 0.00000018, + "input_cost_per_token": 1.8e-07, + "output_cost_per_token": 1.8e-07, "litellm_provider": "together_ai", "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -11202,8 +11830,8 @@ "supports_tool_choice": true }, "together_ai/meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo": { - "input_cost_per_token": 0.00000088, - "output_cost_per_token": 0.00000088, + "input_cost_per_token": 8.8e-07, + "output_cost_per_token": 8.8e-07, "litellm_provider": "together_ai", "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -11212,8 +11840,8 @@ "supports_tool_choice": true }, "together_ai/meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": { - "input_cost_per_token": 0.0000035, - "output_cost_per_token": 0.0000035, + "input_cost_per_token": 3.5e-06, + "output_cost_per_token": 3.5e-06, "litellm_provider": "together_ai", "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -11221,8 +11849,8 @@ "supports_tool_choice": true }, "together_ai/meta-llama/Llama-3.3-70B-Instruct-Turbo": { - "input_cost_per_token": 0.00000088, - "output_cost_per_token": 0.00000088, + "input_cost_per_token": 8.8e-07, + "output_cost_per_token": 8.8e-07, "litellm_provider": "together_ai", "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -11241,8 +11869,8 @@ "supports_tool_choice": true }, "together_ai/mistralai/Mixtral-8x7B-Instruct-v0.1": { - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.0000006, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "together_ai", "supports_function_calling": true, "supports_parallel_function_calling": true, @@ -11315,9 +11943,9 @@ "supports_tool_choice": true }, "ollama/codegemma": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", @@ -11330,7 +11958,7 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "chat", + "mode": "chat", "supports_function_calling": false }, "ollama/deepseek-coder-v2-instruct": { @@ -11340,7 +11968,7 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "chat", + "mode": "chat", "supports_function_calling": true }, "ollama/deepseek-coder-v2-base": { @@ -11350,7 +11978,7 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "completion", + "mode": "completion", "supports_function_calling": true }, "ollama/deepseek-coder-v2-lite-instruct": { @@ -11360,7 +11988,7 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "chat", + "mode": "chat", "supports_function_calling": true }, "ollama/deepseek-coder-v2-lite-base": { @@ -11370,7 +11998,7 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "completion", + "mode": "completion", "supports_function_calling": true }, "ollama/internlm2_5-20b-chat": { @@ -11380,49 +12008,49 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "chat", + "mode": "chat", "supports_function_calling": true }, "ollama/llama2": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", "mode": "chat" }, "ollama/llama2:7b": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", "mode": "chat" }, "ollama/llama2:13b": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", "mode": "chat" }, "ollama/llama2:70b": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", "mode": "chat" }, "ollama/llama2-uncensored": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", @@ -11462,7 +12090,7 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "chat", + "mode": "chat", "supports_function_calling": true }, "ollama/mistral-large-instruct-2407": { @@ -11526,8 +12154,8 @@ "supports_function_calling": true }, "ollama/codellama": { - "max_tokens": 4096, - "max_input_tokens": 4096, + "max_tokens": 4096, + "max_input_tokens": 4096, "max_output_tokens": 4096, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, @@ -11535,8 +12163,8 @@ "mode": "completion" }, "ollama/orca-mini": { - "max_tokens": 4096, - "max_input_tokens": 4096, + "max_tokens": 4096, + "max_input_tokens": 4096, "max_output_tokens": 4096, "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, @@ -11556,8 +12184,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000090, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11566,8 +12194,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000022, - "output_cost_per_token": 0.00000022, + "input_cost_per_token": 2.2e-07, + "output_cost_per_token": 2.2e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11576,8 +12204,8 @@ "max_tokens": 8191, "max_input_tokens": 32768, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000013, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 1.3e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11586,8 +12214,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000090, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11596,8 +12224,8 @@ "max_tokens": 8191, "max_input_tokens": 32768, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000027, - "output_cost_per_token": 0.00000027, + "input_cost_per_token": 2.7e-07, + "output_cost_per_token": 2.7e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11606,8 +12234,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000060, - "output_cost_per_token": 0.00000060, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11616,8 +12244,8 @@ "max_tokens": 4096, "max_input_tokens": 32000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000027, - "output_cost_per_token": 0.00000027, + "input_cost_per_token": 2.7e-07, + "output_cost_per_token": 2.7e-07, "litellm_provider": "deepinfra", "mode": "completion" }, @@ -11625,8 +12253,8 @@ "max_tokens": 4096, "max_input_tokens": 16384, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000060, - "output_cost_per_token": 0.00000060, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11635,8 +12263,8 @@ "max_tokens": 8191, "max_input_tokens": 32768, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000027, - "output_cost_per_token": 0.00000027, + "input_cost_per_token": 2.7e-07, + "output_cost_per_token": 2.7e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11645,8 +12273,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000090, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11655,8 +12283,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000060, - "output_cost_per_token": 0.00000060, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11665,8 +12293,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000013, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 1.3e-07, "litellm_provider": "deepinfra", "mode": "completion" }, @@ -11674,8 +12302,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000090, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11684,8 +12312,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000022, - "output_cost_per_token": 0.00000022, + "input_cost_per_token": 2.2e-07, + "output_cost_per_token": 2.2e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11694,8 +12322,8 @@ "max_tokens": 8191, "max_input_tokens": 32768, "max_output_tokens": 8191, - "input_cost_per_token": 0.00000020, - "output_cost_per_token": 0.00000020, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11704,8 +12332,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000013, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 1.3e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11714,8 +12342,8 @@ "max_tokens": 8191, "max_input_tokens": 8191, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000008, - "output_cost_per_token": 0.00000008, + "input_cost_per_token": 8e-08, + "output_cost_per_token": 8e-08, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11724,8 +12352,8 @@ "max_tokens": 8191, "max_input_tokens": 8191, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000059, - "output_cost_per_token": 0.00000079, + "input_cost_per_token": 5.9e-07, + "output_cost_per_token": 7.9e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -11734,8 +12362,8 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 32768, - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_function_calling": true, @@ -11746,8 +12374,8 @@ "max_tokens": 4096, "max_input_tokens": 200000, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000060, - "output_cost_per_token": 0.00000060, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "deepinfra", "mode": "completion" }, @@ -11755,160 +12383,160 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000013, - "output_cost_per_token": 0.00000013, + "input_cost_per_token": 1.3e-07, + "output_cost_per_token": 1.3e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true }, - "perplexity/codellama-34b-instruct": { + "perplexity/codellama-34b-instruct": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000035, - "output_cost_per_token": 0.00000140, - "litellm_provider": "perplexity", - "mode": "chat" + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 1.4e-06, + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/codellama-70b-instruct": { + "perplexity/codellama-70b-instruct": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000280, - "litellm_provider": "perplexity", - "mode": "chat" + "input_cost_per_token": 7e-07, + "output_cost_per_token": 2.8e-06, + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/llama-3.1-70b-instruct": { + "perplexity/llama-3.1-70b-instruct": { "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000001, - "litellm_provider": "perplexity", - "mode": "chat" + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/llama-3.1-8b-instruct": { + "perplexity/llama-3.1-8b-instruct": { "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, - "litellm_provider": "perplexity", - "mode": "chat" + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/llama-3.1-sonar-huge-128k-online": { + "perplexity/llama-3.1-sonar-huge-128k-online": { "max_tokens": 127072, "max_input_tokens": 127072, "max_output_tokens": 127072, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000005, - "litellm_provider": "perplexity", + "input_cost_per_token": 5e-06, + "output_cost_per_token": 5e-06, + "litellm_provider": "perplexity", "mode": "chat", "deprecation_date": "2025-02-22" }, - "perplexity/llama-3.1-sonar-large-128k-online": { + "perplexity/llama-3.1-sonar-large-128k-online": { "max_tokens": 127072, "max_input_tokens": 127072, "max_output_tokens": 127072, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000001, - "litellm_provider": "perplexity", + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, + "litellm_provider": "perplexity", "mode": "chat", "deprecation_date": "2025-02-22" }, - "perplexity/llama-3.1-sonar-large-128k-chat": { + "perplexity/llama-3.1-sonar-large-128k-chat": { "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000001, - "litellm_provider": "perplexity", + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, + "litellm_provider": "perplexity", "mode": "chat", "deprecation_date": "2025-02-22" }, - "perplexity/llama-3.1-sonar-small-128k-chat": { + "perplexity/llama-3.1-sonar-small-128k-chat": { "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, - "litellm_provider": "perplexity", + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, + "litellm_provider": "perplexity", "mode": "chat", "deprecation_date": "2025-02-22" }, - "perplexity/llama-3.1-sonar-small-128k-online": { + "perplexity/llama-3.1-sonar-small-128k-online": { "max_tokens": 127072, "max_input_tokens": 127072, "max_output_tokens": 127072, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, - "litellm_provider": "perplexity", - "mode": "chat" , + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, + "litellm_provider": "perplexity", + "mode": "chat", "deprecation_date": "2025-02-22" }, - "perplexity/pplx-7b-chat": { + "perplexity/pplx-7b-chat": { "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000007, - "output_cost_per_token": 0.00000028, - "litellm_provider": "perplexity", - "mode": "chat" + "input_cost_per_token": 7e-08, + "output_cost_per_token": 2.8e-07, + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/pplx-70b-chat": { + "perplexity/pplx-70b-chat": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000280, - "litellm_provider": "perplexity", - "mode": "chat" + "input_cost_per_token": 7e-07, + "output_cost_per_token": 2.8e-06, + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/pplx-7b-online": { + "perplexity/pplx-7b-online": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000000, - "output_cost_per_token": 0.00000028, + "input_cost_per_token": 0.0, + "output_cost_per_token": 2.8e-07, "input_cost_per_request": 0.005, - "litellm_provider": "perplexity", - "mode": "chat" + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/pplx-70b-online": { + "perplexity/pplx-70b-online": { "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.0000000, - "output_cost_per_token": 0.00000280, + "max_output_tokens": 4096, + "input_cost_per_token": 0.0, + "output_cost_per_token": 2.8e-06, "input_cost_per_request": 0.005, - "litellm_provider": "perplexity", - "mode": "chat" + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/llama-2-70b-chat": { + "perplexity/llama-2-70b-chat": { "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000280, - "litellm_provider": "perplexity", - "mode": "chat" + "max_output_tokens": 4096, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 2.8e-06, + "litellm_provider": "perplexity", + "mode": "chat" }, - "perplexity/mistral-7b-instruct": { + "perplexity/mistral-7b-instruct": { "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000007, - "output_cost_per_token": 0.00000028, - "litellm_provider": "perplexity", - "mode": "chat" + "max_output_tokens": 4096, + "input_cost_per_token": 7e-08, + "output_cost_per_token": 2.8e-07, + "litellm_provider": "perplexity", + "mode": "chat" }, "perplexity/mixtral-8x7b-instruct": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000007, - "output_cost_per_token": 0.00000028, + "input_cost_per_token": 7e-08, + "output_cost_per_token": 2.8e-07, "litellm_provider": "perplexity", "mode": "chat" }, @@ -11916,8 +12544,8 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000007, - "output_cost_per_token": 0.00000028, + "input_cost_per_token": 7e-08, + "output_cost_per_token": 2.8e-07, "litellm_provider": "perplexity", "mode": "chat" }, @@ -11926,7 +12554,7 @@ "max_input_tokens": 12000, "max_output_tokens": 12000, "input_cost_per_token": 0, - "output_cost_per_token": 0.00000028, + "output_cost_per_token": 2.8e-07, "input_cost_per_request": 0.005, "litellm_provider": "perplexity", "mode": "chat" @@ -11935,8 +12563,8 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.0000018, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 1.8e-06, "litellm_provider": "perplexity", "mode": "chat" }, @@ -11945,7 +12573,7 @@ "max_input_tokens": 12000, "max_output_tokens": 12000, "input_cost_per_token": 0, - "output_cost_per_token": 0.0000018, + "output_cost_per_token": 1.8e-06, "input_cost_per_request": 0.005, "litellm_provider": "perplexity", "mode": "chat" @@ -11953,14 +12581,14 @@ "perplexity/sonar": { "max_tokens": 128000, "max_input_tokens": 128000, - "input_cost_per_token": 1e-6, - "output_cost_per_token": 1e-6, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, "litellm_provider": "perplexity", "mode": "chat", "search_context_cost_per_query": { - "search_context_size_low": 5e-3, - "search_context_size_medium": 8e-3, - "search_context_size_high": 12e-3 + "search_context_size_low": 0.005, + "search_context_size_medium": 0.008, + "search_context_size_high": 0.012 }, "supports_web_search": true }, @@ -11968,28 +12596,28 @@ "max_tokens": 8000, "max_input_tokens": 200000, "max_output_tokens": 8000, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 15e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, "litellm_provider": "perplexity", "mode": "chat", "search_context_cost_per_query": { - "search_context_size_low": 6e-3, - "search_context_size_medium": 10e-3, - "search_context_size_high": 14e-3 + "search_context_size_low": 0.006, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.014 }, "supports_web_search": true }, "perplexity/sonar-reasoning": { "max_tokens": 128000, "max_input_tokens": 128000, - "input_cost_per_token": 1e-6, - "output_cost_per_token": 5e-6, + "input_cost_per_token": 1e-06, + "output_cost_per_token": 5e-06, "litellm_provider": "perplexity", "mode": "chat", "search_context_cost_per_query": { - "search_context_size_low": 5e-3, - "search_context_size_medium": 8e-3, - "search_context_size_high": 14e-3 + "search_context_size_low": 0.005, + "search_context_size_medium": 0.008, + "search_context_size_high": 0.014 }, "supports_web_search": true, "supports_reasoning": true @@ -11997,14 +12625,14 @@ "perplexity/sonar-reasoning-pro": { "max_tokens": 128000, "max_input_tokens": 128000, - "input_cost_per_token": 2e-6, - "output_cost_per_token": 8e-6, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, "litellm_provider": "perplexity", "mode": "chat", "search_context_cost_per_query": { - "search_context_size_low": 6e-3, - "search_context_size_medium": 10e-3, - "search_context_size_high": 14e-3 + "search_context_size_low": 0.006, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.014 }, "supports_web_search": true, "supports_reasoning": true @@ -12012,15 +12640,15 @@ "perplexity/sonar-deep-research": { "max_tokens": 128000, "max_input_tokens": 128000, - "input_cost_per_token": 2e-6, - "output_cost_per_token": 8e-6, - "output_cost_per_reasoning_token": 3e-6, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, + "output_cost_per_reasoning_token": 3e-06, "litellm_provider": "perplexity", "mode": "chat", "search_context_cost_per_query": { - "search_context_size_low": 5e-3, - "search_context_size_medium": 5e-3, - "search_context_size_high": 5e-3 + "search_context_size_low": 0.005, + "search_context_size_medium": 0.005, + "search_context_size_high": 0.005 }, "supports_reasoning": true, "supports_web_search": true @@ -12029,9 +12657,9 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000001, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 1e-07, + "output_cost_per_token": 1e-07, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12042,9 +12670,9 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000001, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 1e-07, + "output_cost_per_token": 1e-07, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12055,9 +12683,9 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000001, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 1e-07, + "output_cost_per_token": 1e-07, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12068,9 +12696,9 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_vision": true, @@ -12082,9 +12710,9 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_vision": true, @@ -12095,9 +12723,9 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12108,9 +12736,9 @@ "max_tokens": 65536, "max_input_tokens": 65536, "max_output_tokens": 65536, - "input_cost_per_token": 0.0000012, - "output_cost_per_token": 0.0000012, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 1.2e-06, + "output_cost_per_token": 1.2e-06, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12121,9 +12749,9 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 32768, - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12134,9 +12762,9 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12147,9 +12775,9 @@ "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 32768, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000003, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 3e-06, + "output_cost_per_token": 3e-06, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12160,9 +12788,9 @@ "max_tokens": 65536, "max_input_tokens": 65536, "max_output_tokens": 65536, - "input_cost_per_token": 0.0000012, - "output_cost_per_token": 0.0000012, - "litellm_provider": "fireworks_ai", + "input_cost_per_token": 1.2e-06, + "output_cost_per_token": 1.2e-06, + "litellm_provider": "fireworks_ai", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, @@ -12173,8 +12801,8 @@ "max_tokens": 8192, "max_input_tokens": 128000, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "fireworks_ai", "mode": "chat", "supports_response_schema": true, @@ -12185,8 +12813,8 @@ "max_tokens": 20480, "max_input_tokens": 128000, "max_output_tokens": 20480, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 8e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 8e-06, "litellm_provider": "fireworks_ai", "mode": "chat", "supports_response_schema": true, @@ -12197,8 +12825,8 @@ "max_tokens": 20480, "max_input_tokens": 128000, "max_output_tokens": 20480, - "input_cost_per_token": 0.55e-6, - "output_cost_per_token": 2.19e-6, + "input_cost_per_token": 5.5e-07, + "output_cost_per_token": 2.19e-06, "litellm_provider": "fireworks_ai", "mode": "chat", "supports_response_schema": true, @@ -12209,8 +12837,8 @@ "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, - "input_cost_per_token": 3e-6, - "output_cost_per_token": 3e-6, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 3e-06, "litellm_provider": "fireworks_ai", "mode": "chat", "supports_response_schema": true, @@ -12221,8 +12849,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.22e-6, - "output_cost_per_token": 0.88e-6, + "input_cost_per_token": 2.2e-07, + "output_cost_per_token": 8.8e-07, "litellm_provider": "fireworks_ai", "mode": "chat", "supports_response_schema": true, @@ -12233,8 +12861,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 0.15e-6, - "output_cost_per_token": 0.60e-6, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "fireworks_ai", "mode": "chat", "supports_response_schema": true, @@ -12244,8 +12872,8 @@ "fireworks_ai/nomic-ai/nomic-embed-text-v1.5": { "max_tokens": 8192, "max_input_tokens": 8192, - "input_cost_per_token": 0.000000008, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 8e-09, + "output_cost_per_token": 0.0, "litellm_provider": "fireworks_ai-embedding-models", "mode": "embedding", "source": "https://fireworks.ai/pricing" @@ -12253,8 +12881,8 @@ "fireworks_ai/nomic-ai/nomic-embed-text-v1": { "max_tokens": 8192, "max_input_tokens": 8192, - "input_cost_per_token": 0.000000008, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 8e-09, + "output_cost_per_token": 0.0, "litellm_provider": "fireworks_ai-embedding-models", "mode": "embedding", "source": "https://fireworks.ai/pricing" @@ -12262,8 +12890,8 @@ "fireworks_ai/WhereIsAI/UAE-Large-V1": { "max_tokens": 512, "max_input_tokens": 512, - "input_cost_per_token": 0.000000016, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.6e-08, + "output_cost_per_token": 0.0, "litellm_provider": "fireworks_ai-embedding-models", "mode": "embedding", "source": "https://fireworks.ai/pricing" @@ -12271,8 +12899,8 @@ "fireworks_ai/thenlper/gte-large": { "max_tokens": 512, "max_input_tokens": 512, - "input_cost_per_token": 0.000000016, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.6e-08, + "output_cost_per_token": 0.0, "litellm_provider": "fireworks_ai-embedding-models", "mode": "embedding", "source": "https://fireworks.ai/pricing" @@ -12280,35 +12908,35 @@ "fireworks_ai/thenlper/gte-base": { "max_tokens": 512, "max_input_tokens": 512, - "input_cost_per_token": 0.000000008, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 8e-09, + "output_cost_per_token": 0.0, "litellm_provider": "fireworks_ai-embedding-models", "mode": "embedding", "source": "https://fireworks.ai/pricing" }, "fireworks-ai-up-to-4b": { - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "fireworks_ai" }, "fireworks-ai-4.1b-to-16b": { - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "fireworks_ai" }, "fireworks-ai-above-16b": { - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, "litellm_provider": "fireworks_ai" }, "fireworks-ai-moe-up-to-56b": { - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000005, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "fireworks_ai" }, "fireworks-ai-56b-to-176b": { - "input_cost_per_token": 0.0000012, - "output_cost_per_token": 0.0000012, + "input_cost_per_token": 1.2e-06, + "output_cost_per_token": 1.2e-06, "litellm_provider": "fireworks_ai" }, "fireworks-ai-default": { @@ -12317,270 +12945,270 @@ "litellm_provider": "fireworks_ai" }, "fireworks-ai-embedding-up-to-150m": { - "input_cost_per_token": 0.000000008, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 8e-09, + "output_cost_per_token": 0.0, "litellm_provider": "fireworks_ai-embedding-models" }, "fireworks-ai-embedding-150m-to-350m": { - "input_cost_per_token": 0.000000016, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.6e-08, + "output_cost_per_token": 0.0, "litellm_provider": "fireworks_ai-embedding-models" }, - "anyscale/mistralai/Mistral-7B-Instruct-v0.1": { + "anyscale/mistralai/Mistral-7B-Instruct-v0.1": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, - "litellm_provider": "anyscale", + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, + "litellm_provider": "anyscale", "mode": "chat", "supports_function_calling": true, "source": "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/mistralai-Mistral-7B-Instruct-v0.1" - }, - "anyscale/mistralai/Mixtral-8x7B-Instruct-v0.1": { + }, + "anyscale/mistralai/Mixtral-8x7B-Instruct-v0.1": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, - "litellm_provider": "anyscale", + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, + "litellm_provider": "anyscale", "mode": "chat", "supports_function_calling": true, "source": "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/mistralai-Mixtral-8x7B-Instruct-v0.1" - }, - "anyscale/mistralai/Mixtral-8x22B-Instruct-v0.1": { + }, + "anyscale/mistralai/Mixtral-8x22B-Instruct-v0.1": { "max_tokens": 65536, "max_input_tokens": 65536, "max_output_tokens": 65536, - "input_cost_per_token": 0.00000090, - "output_cost_per_token": 0.00000090, - "litellm_provider": "anyscale", + "input_cost_per_token": 9e-07, + "output_cost_per_token": 9e-07, + "litellm_provider": "anyscale", "mode": "chat", "supports_function_calling": true, "source": "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/mistralai-Mixtral-8x22B-Instruct-v0.1" - }, - "anyscale/HuggingFaceH4/zephyr-7b-beta": { + }, + "anyscale/HuggingFaceH4/zephyr-7b-beta": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, - "litellm_provider": "anyscale", + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, + "litellm_provider": "anyscale", "mode": "chat" - }, - "anyscale/google/gemma-7b-it": { + }, + "anyscale/google/gemma-7b-it": { "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, - "litellm_provider": "anyscale", + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, + "litellm_provider": "anyscale", "mode": "chat", "source": "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/google-gemma-7b-it" - }, - "anyscale/meta-llama/Llama-2-7b-chat-hf": { + }, + "anyscale/meta-llama/Llama-2-7b-chat-hf": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, - "litellm_provider": "anyscale", + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, + "litellm_provider": "anyscale", "mode": "chat" - }, - "anyscale/meta-llama/Llama-2-13b-chat-hf": { + }, + "anyscale/meta-llama/Llama-2-13b-chat-hf": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000025, - "output_cost_per_token": 0.00000025, - "litellm_provider": "anyscale", + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 2.5e-07, + "litellm_provider": "anyscale", "mode": "chat" - }, - "anyscale/meta-llama/Llama-2-70b-chat-hf": { + }, + "anyscale/meta-llama/Llama-2-70b-chat-hf": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000001, - "litellm_provider": "anyscale", + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, + "litellm_provider": "anyscale", "mode": "chat" - }, - "anyscale/codellama/CodeLlama-34b-Instruct-hf": { + }, + "anyscale/codellama/CodeLlama-34b-Instruct-hf": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000001, - "litellm_provider": "anyscale", + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, + "litellm_provider": "anyscale", "mode": "chat" - }, - "anyscale/codellama/CodeLlama-70b-Instruct-hf": { + }, + "anyscale/codellama/CodeLlama-70b-Instruct-hf": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000001, - "output_cost_per_token": 0.000001, - "litellm_provider": "anyscale", + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, + "litellm_provider": "anyscale", "mode": "chat", - "source" : "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/codellama-CodeLlama-70b-Instruct-hf" - }, - "anyscale/meta-llama/Meta-Llama-3-8B-Instruct": { + "source": "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/codellama-CodeLlama-70b-Instruct-hf" + }, + "anyscale/meta-llama/Meta-Llama-3-8B-Instruct": { "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000015, - "output_cost_per_token": 0.00000015, - "litellm_provider": "anyscale", + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, + "litellm_provider": "anyscale", "mode": "chat", "source": "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/meta-llama-Meta-Llama-3-8B-Instruct" - }, - "anyscale/meta-llama/Meta-Llama-3-70B-Instruct": { + }, + "anyscale/meta-llama/Meta-Llama-3-70B-Instruct": { "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000100, - "output_cost_per_token": 0.00000100, - "litellm_provider": "anyscale", + "input_cost_per_token": 1e-06, + "output_cost_per_token": 1e-06, + "litellm_provider": "anyscale", "mode": "chat", - "source" : "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/meta-llama-Meta-Llama-3-70B-Instruct" - }, - "cloudflare/@cf/meta/llama-2-7b-chat-fp16": { - "max_tokens": 3072, - "max_input_tokens": 3072, - "max_output_tokens": 3072, - "input_cost_per_token": 0.000001923, - "output_cost_per_token": 0.000001923, - "litellm_provider": "cloudflare", + "source": "https://docs.anyscale.com/preview/endpoints/text-generation/supported-models/meta-llama-Meta-Llama-3-70B-Instruct" + }, + "cloudflare/@cf/meta/llama-2-7b-chat-fp16": { + "max_tokens": 3072, + "max_input_tokens": 3072, + "max_output_tokens": 3072, + "input_cost_per_token": 1.923e-06, + "output_cost_per_token": 1.923e-06, + "litellm_provider": "cloudflare", "mode": "chat" - }, - "cloudflare/@cf/meta/llama-2-7b-chat-int8": { - "max_tokens": 2048, - "max_input_tokens": 2048, - "max_output_tokens": 2048, - "input_cost_per_token": 0.000001923, - "output_cost_per_token": 0.000001923, - "litellm_provider": "cloudflare", + }, + "cloudflare/@cf/meta/llama-2-7b-chat-int8": { + "max_tokens": 2048, + "max_input_tokens": 2048, + "max_output_tokens": 2048, + "input_cost_per_token": 1.923e-06, + "output_cost_per_token": 1.923e-06, + "litellm_provider": "cloudflare", "mode": "chat" - }, - "cloudflare/@cf/mistral/mistral-7b-instruct-v0.1": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.000001923, - "output_cost_per_token": 0.000001923, - "litellm_provider": "cloudflare", + }, + "cloudflare/@cf/mistral/mistral-7b-instruct-v0.1": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 1.923e-06, + "output_cost_per_token": 1.923e-06, + "litellm_provider": "cloudflare", "mode": "chat" - }, - "cloudflare/@hf/thebloke/codellama-7b-instruct-awq": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000001923, - "output_cost_per_token": 0.000001923, - "litellm_provider": "cloudflare", - "mode": "chat" - }, - "voyage/voyage-01": { + }, + "cloudflare/@hf/thebloke/codellama-7b-instruct-awq": { "max_tokens": 4096, "max_input_tokens": 4096, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, + "max_output_tokens": 4096, + "input_cost_per_token": 1.923e-06, + "output_cost_per_token": 1.923e-06, + "litellm_provider": "cloudflare", + "mode": "chat" + }, + "voyage/voyage-01": { + "max_tokens": 4096, + "max_input_tokens": 4096, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-lite-01": { "max_tokens": 4096, "max_input_tokens": 4096, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-large-2": { "max_tokens": 16000, "max_input_tokens": 16000, - "input_cost_per_token": 0.00000012, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.2e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-finance-2": { "max_tokens": 32000, "max_input_tokens": 32000, - "input_cost_per_token": 0.00000012, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.2e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-lite-02-instruct": { "max_tokens": 4000, "max_input_tokens": 4000, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-law-2": { "max_tokens": 16000, "max_input_tokens": 16000, - "input_cost_per_token": 0.00000012, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.2e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-code-2": { "max_tokens": 16000, "max_input_tokens": 16000, - "input_cost_per_token": 0.00000012, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.2e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-2": { "max_tokens": 4000, "max_input_tokens": 4000, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-3-large": { "max_tokens": 32000, "max_input_tokens": 32000, - "input_cost_per_token": 0.00000018, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.8e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-3": { "max_tokens": 32000, "max_input_tokens": 32000, - "input_cost_per_token": 0.00000006, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 6e-08, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-3-lite": { "max_tokens": 32000, "max_input_tokens": 32000, - "input_cost_per_token": 0.00000002, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 2e-08, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-code-3": { "max_tokens": 32000, "max_input_tokens": 32000, - "input_cost_per_token": 0.00000018, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.8e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, "voyage/voyage-multimodal-3": { "max_tokens": 32000, "max_input_tokens": 32000, - "input_cost_per_token": 0.00000012, - "output_cost_per_token": 0.000000, + "input_cost_per_token": 1.2e-07, + "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "embedding" }, @@ -12589,8 +13217,8 @@ "max_input_tokens": 16000, "max_output_tokens": 16000, "max_query_tokens": 16000, - "input_cost_per_token": 0.00000005, - "input_cost_per_query": 0.00000005, + "input_cost_per_token": 5e-08, + "input_cost_per_query": 5e-08, "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "rerank" @@ -12600,8 +13228,8 @@ "max_input_tokens": 8000, "max_output_tokens": 8000, "max_query_tokens": 8000, - "input_cost_per_token": 0.00000002, - "input_cost_per_query": 0.00000002, + "input_cost_per_token": 2e-08, + "input_cost_per_query": 2e-08, "output_cost_per_token": 0.0, "litellm_provider": "voyage", "mode": "rerank" @@ -12609,15 +13237,17 @@ "databricks/databricks-claude-3-7-sonnet": { "max_tokens": 200000, "max_input_tokens": 200000, - "max_output_tokens": 128000, - "input_cost_per_token": 0.0000025, - "input_dbu_cost_per_token": 0.00003571, - "output_cost_per_token": 0.000017857, + "max_output_tokens": 128000, + "input_cost_per_token": 2.5e-06, + "input_dbu_cost_per_token": 3.571e-05, + "output_cost_per_token": 1.7857e-05, "output_db_cost_per_token": 0.000214286, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Claude 3.7 conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Claude 3.7 conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_assistant_prefill": true, "supports_function_calling": true, "supports_tool_choice": true, @@ -12626,175 +13256,199 @@ "databricks/databricks-meta-llama-3-1-405b-instruct": { "max_tokens": 128000, "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 0.000005, - "input_dbu_cost_per_token": 0.000071429, - "output_cost_per_token": 0.00001500002, + "max_output_tokens": 128000, + "input_cost_per_token": 5e-06, + "input_dbu_cost_per_token": 7.1429e-05, + "output_cost_per_token": 1.500002e-05, "output_db_cost_per_token": 0.000214286, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-meta-llama-3-1-70b-instruct": { "max_tokens": 128000, "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 0.00000100002, - "input_dbu_cost_per_token": 0.000014286, - "output_cost_per_token": 0.00000299999, - "output_dbu_cost_per_token": 0.000042857, + "max_output_tokens": 128000, + "input_cost_per_token": 1.00002e-06, + "input_dbu_cost_per_token": 1.4286e-05, + "output_cost_per_token": 2.99999e-06, + "output_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-meta-llama-3-3-70b-instruct": { "max_tokens": 128000, "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 0.00000100002, - "input_dbu_cost_per_token": 0.000014286, - "output_cost_per_token": 0.00000299999, - "output_dbu_cost_per_token": 0.000042857, + "max_output_tokens": 128000, + "input_cost_per_token": 1.00002e-06, + "input_dbu_cost_per_token": 1.4286e-05, + "output_cost_per_token": 2.99999e-06, + "output_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-llama-4-maverick": { "max_tokens": 128000, "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 0.000005, - "input_dbu_cost_per_token": 0.00007143, - "output_cost_per_token": 0.000015, + "max_output_tokens": 128000, + "input_cost_per_token": 5e-06, + "input_dbu_cost_per_token": 7.143e-05, + "output_cost_per_token": 1.5e-05, "output_dbu_cost_per_token": 0.00021429, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Databricks documentation now provides both DBU costs (_dbu_cost_per_token) and dollar costs(_cost_per_token)."}, + "metadata": { + "notes": "Databricks documentation now provides both DBU costs (_dbu_cost_per_token) and dollar costs(_cost_per_token)." + }, "supports_tool_choice": true }, "databricks/databricks-dbrx-instruct": { "max_tokens": 32768, "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 0.00000074998, - "input_dbu_cost_per_token": 0.000010714, - "output_cost_per_token": 0.00000224901, - "output_dbu_cost_per_token": 0.000032143, + "max_output_tokens": 32768, + "input_cost_per_token": 7.4998e-07, + "input_dbu_cost_per_token": 1.0714e-05, + "output_cost_per_token": 2.24901e-06, + "output_dbu_cost_per_token": 3.2143e-05, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-meta-llama-3-70b-instruct": { "max_tokens": 128000, "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 0.00000100002, - "input_dbu_cost_per_token": 0.000014286, - "output_cost_per_token": 0.00000299999, - "output_dbu_cost_per_token": 0.000042857, + "max_output_tokens": 128000, + "input_cost_per_token": 1.00002e-06, + "input_dbu_cost_per_token": 1.4286e-05, + "output_cost_per_token": 2.99999e-06, + "output_dbu_cost_per_token": 4.2857e-05, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-llama-2-70b-chat": { "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000050001, - "input_dbu_cost_per_token": 0.000007143, - "output_cost_per_token": 0.0000015, - "output_dbu_cost_per_token": 0.000021429, + "max_output_tokens": 4096, + "input_cost_per_token": 5.0001e-07, + "input_dbu_cost_per_token": 7.143e-06, + "output_cost_per_token": 1.5e-06, + "output_dbu_cost_per_token": 2.1429e-05, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-mixtral-8x7b-instruct": { "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000050001, - "input_dbu_cost_per_token": 0.000007143, - "output_cost_per_token": 0.00000099902, - "output_dbu_cost_per_token": 0.000014286, + "max_output_tokens": 4096, + "input_cost_per_token": 5.0001e-07, + "input_dbu_cost_per_token": 7.143e-06, + "output_cost_per_token": 9.9902e-07, + "output_dbu_cost_per_token": 1.4286e-05, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-mpt-30b-instruct": { "max_tokens": 8192, "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000099902, - "input_dbu_cost_per_token": 0.000014286, - "output_cost_per_token": 0.00000099902, - "output_dbu_cost_per_token": 0.000014286, + "max_output_tokens": 8192, + "input_cost_per_token": 9.9902e-07, + "input_dbu_cost_per_token": 1.4286e-05, + "output_cost_per_token": 9.9902e-07, + "output_dbu_cost_per_token": 1.4286e-05, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-mpt-7b-instruct": { "max_tokens": 8192, "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000050001, - "input_dbu_cost_per_token": 0.000007143, + "max_output_tokens": 8192, + "input_cost_per_token": 5.0001e-07, + "input_dbu_cost_per_token": 7.143e-06, "output_cost_per_token": 0.0, "output_dbu_cost_per_token": 0.0, "litellm_provider": "databricks", "mode": "chat", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."}, + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + }, "supports_tool_choice": true }, "databricks/databricks-bge-large-en": { "max_tokens": 512, "max_input_tokens": 512, - "output_vector_size": 1024, - "input_cost_per_token": 0.00000010003, - "input_dbu_cost_per_token": 0.000001429, + "output_vector_size": 1024, + "input_cost_per_token": 1.0003e-07, + "input_dbu_cost_per_token": 1.429e-06, "output_cost_per_token": 0.0, "output_dbu_cost_per_token": 0.0, "litellm_provider": "databricks", "mode": "embedding", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."} + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + } }, "databricks/databricks-gte-large-en": { "max_tokens": 8192, "max_input_tokens": 8192, - "output_vector_size": 1024, - "input_cost_per_token": 0.00000012999, - "input_dbu_cost_per_token": 0.000001857, + "output_vector_size": 1024, + "input_cost_per_token": 1.2999e-07, + "input_dbu_cost_per_token": 1.857e-06, "output_cost_per_token": 0.0, "output_dbu_cost_per_token": 0.0, "litellm_provider": "databricks", "mode": "embedding", "source": "https://www.databricks.com/product/pricing/foundation-model-serving", - "metadata": {"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."} + "metadata": { + "notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation." + } }, "sambanova/Meta-Llama-3.1-8B-Instruct": { "max_tokens": 16384, "max_input_tokens": 16384, - "max_output_tokens": 16384, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000002, + "max_output_tokens": 16384, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "sambanova", "mode": "chat", "supports_function_calling": true, @@ -12805,9 +13459,9 @@ "sambanova/Meta-Llama-3.1-405B-Instruct": { "max_tokens": 16384, "max_input_tokens": 16384, - "max_output_tokens": 16384, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000010, + "max_output_tokens": 16384, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "sambanova", "mode": "chat", "supports_function_calling": true, @@ -12818,9 +13472,9 @@ "sambanova/Meta-Llama-3.2-1B-Instruct": { "max_tokens": 16384, "max_input_tokens": 16384, - "max_output_tokens": 16384, - "input_cost_per_token": 0.00000004, - "output_cost_per_token": 0.00000008, + "max_output_tokens": 16384, + "input_cost_per_token": 4e-08, + "output_cost_per_token": 8e-08, "litellm_provider": "sambanova", "mode": "chat", "source": "https://cloud.sambanova.ai/plans/pricing" @@ -12828,9 +13482,9 @@ "sambanova/Meta-Llama-3.2-3B-Instruct": { "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000008, - "output_cost_per_token": 0.00000016, + "max_output_tokens": 4096, + "input_cost_per_token": 8e-08, + "output_cost_per_token": 1.6e-07, "litellm_provider": "sambanova", "mode": "chat", "source": "https://cloud.sambanova.ai/plans/pricing" @@ -12838,9 +13492,9 @@ "sambanova/Llama-4-Maverick-17B-128E-Instruct": { "max_tokens": 131072, "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 0.00000063, - "output_cost_per_token": 0.0000018, + "max_output_tokens": 131072, + "input_cost_per_token": 6.3e-07, + "output_cost_per_token": 1.8e-06, "litellm_provider": "sambanova", "mode": "chat", "supports_function_calling": true, @@ -12848,28 +13502,32 @@ "supports_response_schema": true, "supports_vision": true, "source": "https://cloud.sambanova.ai/plans/pricing", - "metadata": {"notes": "For vision models, images are converted to 6432 input tokens and are billed at that amount"} + "metadata": { + "notes": "For vision models, images are converted to 6432 input tokens and are billed at that amount" + } }, "sambanova/Llama-4-Scout-17B-16E-Instruct": { "max_tokens": 8192, "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.0000004, - "output_cost_per_token": 0.0000007, + "max_output_tokens": 8192, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 7e-07, "litellm_provider": "sambanova", "mode": "chat", "supports_function_calling": true, "supports_tool_choice": true, "supports_response_schema": true, "source": "https://cloud.sambanova.ai/plans/pricing", - "metadata": {"notes": "For vision models, images are converted to 6432 input tokens and are billed at that amount"} + "metadata": { + "notes": "For vision models, images are converted to 6432 input tokens and are billed at that amount" + } }, "sambanova/Meta-Llama-3.3-70B-Instruct": { "max_tokens": 131072, "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.0000012, + "max_output_tokens": 131072, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 1.2e-06, "litellm_provider": "sambanova", "mode": "chat", "supports_function_calling": true, @@ -12880,9 +13538,9 @@ "sambanova/Meta-Llama-Guard-3-8B": { "max_tokens": 16384, "max_input_tokens": 16384, - "max_output_tokens": 16384, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000003, + "max_output_tokens": 16384, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 3e-07, "litellm_provider": "sambanova", "mode": "chat", "source": "https://cloud.sambanova.ai/plans/pricing" @@ -12890,9 +13548,9 @@ "sambanova/Qwen3-32B": { "max_tokens": 8192, "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 0.0000004, - "output_cost_per_token": 0.0000008, + "max_output_tokens": 8192, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 8e-07, "litellm_provider": "sambanova", "supports_function_calling": true, "supports_tool_choice": true, @@ -12903,9 +13561,9 @@ "sambanova/QwQ-32B": { "max_tokens": 16384, "max_input_tokens": 16384, - "max_output_tokens": 16384, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.0000010, + "max_output_tokens": 16384, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1e-06, "litellm_provider": "sambanova", "mode": "chat", "source": "https://cloud.sambanova.ai/plans/pricing" @@ -12913,8 +13571,8 @@ "sambanova/Qwen2-Audio-7B-Instruct": { "max_tokens": 4096, "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 0.0000005, + "max_output_tokens": 4096, + "input_cost_per_token": 5e-07, "output_cost_per_token": 0.0001, "litellm_provider": "sambanova", "mode": "chat", @@ -12924,9 +13582,9 @@ "sambanova/DeepSeek-R1-Distill-Llama-70B": { "max_tokens": 131072, "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 0.0000007, - "output_cost_per_token": 0.0000014, + "max_output_tokens": 131072, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 1.4e-06, "litellm_provider": "sambanova", "mode": "chat", "source": "https://cloud.sambanova.ai/plans/pricing" @@ -12934,9 +13592,9 @@ "sambanova/DeepSeek-R1": { "max_tokens": 32768, "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000007, + "max_output_tokens": 32768, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 7e-06, "litellm_provider": "sambanova", "mode": "chat", "source": "https://cloud.sambanova.ai/plans/pricing" @@ -12944,9 +13602,9 @@ "sambanova/DeepSeek-V3-0324": { "max_tokens": 32768, "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 0.0000030, - "output_cost_per_token": 0.0000045, + "max_output_tokens": 32768, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 4.5e-06, "litellm_provider": "sambanova", "mode": "chat", "supports_function_calling": true, @@ -12957,13 +13615,13 @@ "assemblyai/nano": { "mode": "audio_transcription", "input_cost_per_second": 0.00010278, - "output_cost_per_second": 0.00, + "output_cost_per_second": 0.0, "litellm_provider": "assemblyai" }, "assemblyai/best": { "mode": "audio_transcription", - "input_cost_per_second": 0.00003333, - "output_cost_per_second": 0.00, + "input_cost_per_second": 3.333e-05, + "output_cost_per_second": 0.0, "litellm_provider": "assemblyai" }, "jina-reranker-v2-base-multilingual": { @@ -12971,8 +13629,8 @@ "max_input_tokens": 1024, "max_output_tokens": 1024, "max_document_chunks_per_query": 2048, - "input_cost_per_token": 0.000000018, - "output_cost_per_token": 0.000000018, + "input_cost_per_token": 1.8e-08, + "output_cost_per_token": 1.8e-08, "litellm_provider": "jina_ai", "mode": "rerank" }, @@ -13147,43 +13805,43 @@ "mode": "chat" }, "nscale/meta-llama/Llama-4-Scout-17B-16E-Instruct": { - "input_cost_per_token": 9e-8, - "output_cost_per_token": 2.9e-7, + "input_cost_per_token": 9e-08, + "output_cost_per_token": 2.9e-07, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" }, "nscale/Qwen/Qwen2.5-Coder-3B-Instruct": { - "input_cost_per_token": 1e-8, - "output_cost_per_token": 3e-8, + "input_cost_per_token": 1e-08, + "output_cost_per_token": 3e-08, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" }, "nscale/Qwen/Qwen2.5-Coder-7B-Instruct": { - "input_cost_per_token": 1e-8, - "output_cost_per_token": 3e-8, + "input_cost_per_token": 1e-08, + "output_cost_per_token": 3e-08, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" }, "nscale/Qwen/Qwen2.5-Coder-32B-Instruct": { - "input_cost_per_token": 6e-8, - "output_cost_per_token": 2e-7, + "input_cost_per_token": 6e-08, + "output_cost_per_token": 2e-07, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" }, "nscale/Qwen/QwQ-32B": { - "input_cost_per_token": 1.8e-7, - "output_cost_per_token": 2e-7, + "input_cost_per_token": 1.8e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models" }, "nscale/deepseek-ai/DeepSeek-R1-Distill-Llama-70B": { - "input_cost_per_token": 3.75e-7, - "output_cost_per_token": 3.75e-7, + "input_cost_per_token": 3.75e-07, + "output_cost_per_token": 3.75e-07, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13192,8 +13850,8 @@ } }, "nscale/deepseek-ai/DeepSeek-R1-Distill-Llama-8B": { - "input_cost_per_token": 2.5e-8, - "output_cost_per_token": 2.5e-8, + "input_cost_per_token": 2.5e-08, + "output_cost_per_token": 2.5e-08, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13202,8 +13860,8 @@ } }, "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": { - "input_cost_per_token": 9e-8, - "output_cost_per_token": 9e-8, + "input_cost_per_token": 9e-08, + "output_cost_per_token": 9e-08, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13212,8 +13870,8 @@ } }, "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": { - "input_cost_per_token": 2e-7, - "output_cost_per_token": 2e-7, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13222,8 +13880,8 @@ } }, "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": { - "input_cost_per_token": 7e-8, - "output_cost_per_token": 7e-8, + "input_cost_per_token": 7e-08, + "output_cost_per_token": 7e-08, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13232,8 +13890,8 @@ } }, "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": { - "input_cost_per_token": 1.5e-7, - "output_cost_per_token": 1.5e-7, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 1.5e-07, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13242,8 +13900,8 @@ } }, "nscale/mistralai/mixtral-8x22b-instruct-v0.1": { - "input_cost_per_token": 6e-7, - "output_cost_per_token": 6e-7, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 6e-07, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13252,8 +13910,8 @@ } }, "nscale/meta-llama/Llama-3.1-8B-Instruct": { - "input_cost_per_token": 3e-8, - "output_cost_per_token": 3e-8, + "input_cost_per_token": 3e-08, + "output_cost_per_token": 3e-08, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13262,8 +13920,8 @@ } }, "nscale/meta-llama/Llama-3.3-70B-Instruct": { - "input_cost_per_token": 2e-7, - "output_cost_per_token": 2e-7, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "nscale", "mode": "chat", "source": "https://docs.nscale.com/docs/inference/serverless-models/current#chat-models", @@ -13273,7 +13931,7 @@ }, "nscale/black-forest-labs/FLUX.1-schnell": { "mode": "image_generation", - "input_cost_per_pixel": 1.3e-9, + "input_cost_per_pixel": 1.3e-09, "output_cost_per_pixel": 0.0, "litellm_provider": "nscale", "supported_endpoints": [ @@ -13283,7 +13941,7 @@ }, "nscale/stabilityai/stable-diffusion-xl-base-1.0": { "mode": "image_generation", - "input_cost_per_pixel": 3e-9, + "input_cost_per_pixel": 3e-09, "output_cost_per_pixel": 0.0, "litellm_provider": "nscale", "supported_endpoints": [ @@ -13305,4 +13963,4 @@ "litellm_provider": "featherless_ai", "mode": "chat" } -} +} \ No newline at end of file