diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index f5ce53131eb..e99e3540bb1 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -3,9 +3,9 @@ "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, + "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, @@ -19,10 +19,23 @@ "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 }, + "file_search_cost_per_1k_calls": 0.0, + "file_search_cost_per_gb_per_day": 0.0, + "vector_store_cost_per_gb_per_day": 0.0, + "computer_use_input_cost_per_1k_tokens": 0.0, + "computer_use_output_cost_per_1k_tokens": 0.0, + "code_interpreter_cost_per_session": 0.0, + "supported_regions": [ + "global", + "us-west-2", + "eu-west-1", + "ap-southeast-1", + "ap-northeast-1" + ], "deprecation_date": "date when the model becomes deprecated in the format YYYY-MM-DD" }, "omni-moderation-latest": { @@ -56,8 +69,8 @@ "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, @@ -69,16 +82,26 @@ "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, "supports_response_schema": true, @@ -86,28 +109,32 @@ "supports_prompt_caching": true, "supports_system_messages": true, "supports_tool_choice": true, - "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 - } + "supports_native_streaming": true }, "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, "supports_response_schema": true, @@ -115,28 +142,32 @@ "supports_prompt_caching": true, "supports_system_messages": true, "supports_tool_choice": true, - "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 - } + "supports_native_streaming": true }, "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, "supports_response_schema": true, @@ -144,28 +175,32 @@ "supports_prompt_caching": true, "supports_system_messages": true, "supports_tool_choice": true, - "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 - } + "supports_native_streaming": true }, "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, "supports_response_schema": true, @@ -173,28 +208,32 @@ "supports_prompt_caching": true, "supports_system_messages": true, "supports_tool_choice": true, - "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 - } + "supports_native_streaming": true }, "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, "supports_response_schema": true, @@ -208,16 +247,26 @@ "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, "supports_response_schema": true, @@ -231,26 +280,21 @@ "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, "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": 0.030, - "search_context_size_medium": 0.035, - "search_context_size_high": 0.050 - } + "supports_tool_choice": true }, "watsonx/ibm/granite-3-8b-instruct": { "max_tokens": 8192, @@ -274,38 +318,34 @@ "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, "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": 0.030, - "search_context_size_medium": 0.035, - "search_context_size_high": 0.050 - } - }, + "supports_tool_choice": true + }, "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, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -315,22 +355,23 @@ "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, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -343,28 +384,30 @@ "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, "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_tool_choice": true, + "deprecation_date": "2025-07-14" }, "gpt-4o-audio-preview": { "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", @@ -379,10 +422,10 @@ "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, @@ -396,9 +439,9 @@ "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", @@ -409,14 +452,48 @@ "supports_system_messages": true, "supports_tool_choice": true }, + "gpt-4o-audio-preview-2025-06-03": { + "max_tokens": 16384, + "max_input_tokens": 128000, + "max_output_tokens": 16384, + "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, + "supports_parallel_function_calling": true, + "supports_audio_input": true, + "supports_audio_output": true, + "supports_system_messages": true, + "supports_tool_choice": true + }, + "gpt-4o-mini-audio-preview": { + "max_tokens": 16384, + "max_input_tokens": 128000, + "max_output_tokens": 16384, + "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, + "supports_parallel_function_calling": true, + "supports_audio_input": true, + "supports_audio_output": true, + "supports_system_messages": true, + "supports_tool_choice": true + }, "gpt-4o-mini-audio-preview-2024-12-17": { "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, @@ -430,63 +507,54 @@ "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, "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 - } + "supports_tool_choice": true }, - "gpt-4o-mini-search-preview-2025-03-11":{ + "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, "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 - } + "supports_tool_choice": true }, "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, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -498,20 +566,21 @@ "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, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -520,21 +589,51 @@ "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 } }, + "codex-mini-latest": { + "max_tokens": 100000, + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 6e-06, + "cache_read_input_token_cost": 3.75e-07, + "litellm_provider": "openai", + "mode": "responses", + "supports_pdf_input": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supported_endpoints": [ + "/v1/responses" + ] + }, "o1-pro": { "max_tokens": 100000, "max_input_tokens": 200000, "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", + "supports_pdf_input": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_vision": true, @@ -544,9 +643,17 @@ "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, @@ -554,10 +661,11 @@ "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", + "supports_pdf_input": true, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_vision": true, @@ -567,22 +675,31 @@ "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, "supports_parallel_function_calling": true, "supports_vision": true, + "supports_pdf_input": true, "supports_prompt_caching": true, "supports_system_messages": true, "supports_response_schema": true, @@ -593,55 +710,237 @@ "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, + "supports_pdf_input": true, "supports_prompt_caching": true }, + "computer-use-preview": { + "max_tokens": 1024, + "max_input_tokens": 8192, + "max_output_tokens": 1024, + "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" + ], + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_response_schema": true, + "supports_vision": true, + "supports_prompt_caching": false, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_reasoning": true + }, + "o3-deep-research": { + "max_tokens": 100000, + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "input_cost_per_token": 10e-06, + "output_cost_per_token": 40e-06, + "input_cost_per_token_batches": 5e-06, + "output_cost_per_token_batches": 20e-06, + "cache_read_input_token_cost": 2.5e-06, + "litellm_provider": "openai", + "mode": "responses", + "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, + "supports_response_schema": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_native_streaming": true + }, + "o3-deep-research-2025-06-26": { + "max_tokens": 100000, + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "input_cost_per_token": 10e-06, + "output_cost_per_token": 40e-06, + "input_cost_per_token_batches": 5e-06, + "output_cost_per_token_batches": 20e-06, + "cache_read_input_token_cost": 2.5e-06, + "litellm_provider": "openai", + "mode": "responses", + "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, + "supports_response_schema": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_native_streaming": true + }, + "o3-pro": { + "max_tokens": 100000, + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "input_cost_per_token": 2e-05, + "input_cost_per_token_batches": 1e-05, + "output_cost_per_token_batches": 4e-05, + "output_cost_per_token": 8e-05, + "litellm_provider": "openai", + "mode": "responses", + "supports_function_calling": true, + "supports_parallel_function_calling": false, + "supports_vision": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supported_endpoints": [ + "/v1/responses", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ] + }, + "o3-pro-2025-06-10": { + "max_tokens": 100000, + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "input_cost_per_token": 2e-05, + "input_cost_per_token_batches": 1e-05, + "output_cost_per_token_batches": 4e-05, + "output_cost_per_token": 8e-05, + "litellm_provider": "openai", + "mode": "responses", + "supports_function_calling": true, + "supports_parallel_function_calling": false, + "supports_vision": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supported_endpoints": [ + "/v1/responses", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ] + }, "o3": { "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": 2e-06, + "output_cost_per_token": 8e-06, + "cache_read_input_token_cost": 5e-07, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, "supports_parallel_function_calling": false, "supports_vision": true, + "supports_pdf_input": true, "supports_prompt_caching": true, "supports_response_schema": true, "supports_reasoning": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supported_endpoints": [ + "/v1/responses", + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ] }, "o3-2025-04-16": { "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": 2e-06, + "output_cost_per_token": 8e-06, + "cache_read_input_token_cost": 5e-07, "litellm_provider": "openai", "mode": "chat", "supports_function_calling": true, "supports_parallel_function_calling": false, "supports_vision": true, + "supports_pdf_input": true, "supports_prompt_caching": true, "supports_response_schema": true, "supports_reasoning": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supported_endpoints": [ + "/v1/responses", + "/v1/chat/completions", + "/v1/completions", + "/v1/batch" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ] }, "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": "openai", "mode": "chat", "supports_function_calling": true, @@ -656,9 +955,9 @@ "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, @@ -673,11 +972,12 @@ "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, "supports_function_calling": true, "supports_parallel_function_calling": false, "supports_vision": true, @@ -686,15 +986,82 @@ "supports_reasoning": true, "supports_tool_choice": true }, + "o4-mini-deep-research": { + "max_tokens": 100000, + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "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": 0.5e-06, + "litellm_provider": "openai", + "mode": "responses", + "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, + "supports_response_schema": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_native_streaming": true + }, + "o4-mini-deep-research-2025-06-26": { + "max_tokens": 100000, + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "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": 0.5e-06, + "litellm_provider": "openai", + "mode": "responses", + "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, + "supports_response_schema": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_native_streaming": true + }, "o4-mini-2025-04-16": { "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, "supports_function_calling": true, "supports_parallel_function_calling": false, "supports_vision": true, @@ -707,11 +1074,12 @@ "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, "supports_vision": true, "supports_reasoning": true, "supports_prompt_caching": true @@ -720,11 +1088,12 @@ "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, "supports_vision": true, "supports_reasoning": true, "supports_prompt_caching": true @@ -733,11 +1102,12 @@ "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, "supports_vision": true, "supports_reasoning": true, "supports_prompt_caching": true @@ -746,11 +1116,12 @@ "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, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_vision": true, @@ -764,10 +1135,11 @@ "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, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_vision": true, @@ -779,12 +1151,13 @@ "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, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_vision": true, @@ -796,38 +1169,34 @@ "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, "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": 0.030, - "search_context_size_medium": 0.035, - "search_context_size_high": 0.050 - } + "supports_tool_choice": true }, "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, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -840,11 +1209,11 @@ "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", @@ -859,11 +1228,11 @@ "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, @@ -877,11 +1246,11 @@ "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, @@ -895,12 +1264,12 @@ "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, @@ -914,12 +1283,12 @@ "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, @@ -933,10 +1302,11 @@ "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, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_prompt_caching": true, @@ -947,8 +1317,8 @@ "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, @@ -959,8 +1329,8 @@ "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, @@ -973,7 +1343,7 @@ "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", @@ -985,7 +1355,7 @@ "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", @@ -997,7 +1367,7 @@ "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", @@ -1009,10 +1379,11 @@ "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, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_vision": true, @@ -1024,10 +1395,11 @@ "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, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_vision": true, @@ -1039,8 +1411,8 @@ "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, @@ -1053,8 +1425,8 @@ "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, @@ -1067,11 +1439,12 @@ "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, + "supports_pdf_input": true, "supports_prompt_caching": true, "supports_system_messages": true, "deprecation_date": "2024-12-06", @@ -1081,11 +1454,12 @@ "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, + "supports_pdf_input": true, "supports_prompt_caching": true, "supports_system_messages": true, "deprecation_date": "2024-12-06", @@ -1095,8 +1469,8 @@ "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, @@ -1108,8 +1482,8 @@ "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, @@ -1120,8 +1494,8 @@ "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, @@ -1133,8 +1507,8 @@ "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, @@ -1147,8 +1521,8 @@ "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, @@ -1161,8 +1535,8 @@ "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, @@ -1173,8 +1547,8 @@ "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, @@ -1185,10 +1559,10 @@ "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, @@ -1198,8 +1572,8 @@ "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, @@ -1209,8 +1583,8 @@ "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, @@ -1220,8 +1594,8 @@ "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, @@ -1231,8 +1605,8 @@ "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, @@ -1244,12 +1618,13 @@ "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, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -1261,11 +1636,12 @@ "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, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_response_schema": true, @@ -1278,17 +1654,18 @@ "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, "supports_parallel_function_calling": true, "supports_response_schema": true, "supports_vision": true, + "supports_pdf_input": true, "supports_prompt_caching": true, "supports_system_messages": true, "supports_tool_choice": true @@ -1297,10 +1674,10 @@ "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" }, @@ -1308,10 +1685,10 @@ "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" }, @@ -1319,10 +1696,10 @@ "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" }, @@ -1330,10 +1707,10 @@ "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, + "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" }, @@ -1341,18 +1718,18 @@ "max_tokens": 8191, "max_input_tokens": 8191, "output_vector_size": 1536, - "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": "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" }, @@ -1360,8 +1737,8 @@ "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" }, @@ -1369,8 +1746,8 @@ "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" }, @@ -1378,181 +1755,258 @@ "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", - "input_cost_per_token": 0.0000025, - "input_cost_per_audio_token": 0.000006, - "output_cost_per_token": 0.00001, + "max_input_tokens": 16000, + "max_output_tokens": 2000, + "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", - "input_cost_per_token": 0.00000125, - "input_cost_per_audio_token": 0.000003, - "output_cost_per_token": 0.000005, + "max_input_tokens": 16000, + "max_output_tokens": 2000, + "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, "litellm_provider": "openai", - "supported_endpoints": ["/v1/audio/transcriptions"] + "supported_endpoints": [ + "/v1/audio/transcriptions" + ] }, "tts-1": { "mode": "audio_speech", - "input_cost_per_character": 0.000015, + "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, + "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-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" + ] + }, + "azure/gpt-4o-mini-tts": { + "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" + ] }, "azure/computer-use-preview": { "max_tokens": 1024, "max_input_tokens": 8192, "max_output_tokens": 1024, - "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": "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, @@ -1566,15 +2020,23 @@ "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, @@ -1589,15 +2051,23 @@ "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, @@ -1612,16 +2082,25 @@ "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, @@ -1632,25 +2111,34 @@ "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, @@ -1661,25 +2149,34 @@ "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, @@ -1690,25 +2187,34 @@ "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, @@ -1719,25 +2225,34 @@ "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, @@ -1751,16 +2266,25 @@ "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, @@ -1770,18 +2294,87 @@ "supports_tool_choice": true, "supports_native_streaming": true }, + "azure/o3-pro": { + "max_tokens": 100000, + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "input_cost_per_token": 2e-05, + "output_cost_per_token": 8e-05, + "input_cost_per_token_batches": 1e-05, + "output_cost_per_token_batches": 4e-05, + "litellm_provider": "azure", + "mode": "responses", + "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, + "supports_prompt_caching": false, + "supports_response_schema": true, + "supports_reasoning": true, + "supports_tool_choice": true + }, + "azure/o3-pro-2025-06-10": { + "max_tokens": 100000, + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "input_cost_per_token": 2e-05, + "output_cost_per_token": 8e-05, + "input_cost_per_token_batches": 1e-05, + "output_cost_per_token_batches": 4e-05, + "litellm_provider": "azure", + "mode": "responses", + "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, + "supports_prompt_caching": false, + "supports_response_schema": true, + "supports_reasoning": true, + "supports_tool_choice": true + }, "azure/o3": { "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": 2e-06, + "output_cost_per_token": 8e-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": false, "supports_vision": true, @@ -1794,14 +2387,23 @@ "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, @@ -1814,14 +2416,23 @@ "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, @@ -1834,12 +2445,12 @@ "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, @@ -1853,12 +2464,12 @@ "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, @@ -1872,12 +2483,12 @@ "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, @@ -1891,15 +2502,21 @@ "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, @@ -1911,16 +2528,22 @@ "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, @@ -1932,16 +2555,22 @@ "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, @@ -1953,11 +2582,11 @@ "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", @@ -1972,11 +2601,11 @@ "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", @@ -1991,11 +2620,11 @@ "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", @@ -2010,9 +2639,9 @@ "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, @@ -2027,9 +2656,9 @@ "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, @@ -2041,11 +2670,11 @@ "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, @@ -2057,11 +2686,11 @@ "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, @@ -2071,12 +2700,12 @@ }, "azure/tts-1": { "mode": "audio_speech", - "input_cost_per_character": 0.000015, + "input_cost_per_character": 1.5e-05, "litellm_provider": "azure" }, "azure/tts-1-hd": { "mode": "audio_speech", - "input_cost_per_character": 0.000030, + "input_cost_per_character": 3e-05, "litellm_provider": "azure" }, "azure/whisper-1": { @@ -2085,13 +2714,37 @@ "output_cost_per_second": 0.0001, "litellm_provider": "azure" }, + "azure/gpt-4o-transcribe": { + "mode": "audio_transcription", + "max_input_tokens": 16000, + "max_output_tokens": 2000, + "input_cost_per_token": 2.5e-06, + "input_cost_per_audio_token": 6e-06, + "output_cost_per_token": 1e-05, + "litellm_provider": "azure", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ] + }, + "azure/gpt-4o-mini-transcribe": { + "mode": "audio_transcription", + "max_input_tokens": 16000, + "max_output_tokens": 2000, + "input_cost_per_token": 1.25e-06, + "input_cost_per_audio_token": 3e-06, + "output_cost_per_token": 5e-06, + "litellm_provider": "azure", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ] + }, "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, @@ -2104,9 +2757,9 @@ "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, @@ -2119,9 +2772,9 @@ "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": 1.1e-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, @@ -2134,11 +2787,11 @@ "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, @@ -2150,11 +2803,11 @@ "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, @@ -2166,9 +2819,9 @@ "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, @@ -2182,9 +2835,9 @@ "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, @@ -2198,9 +2851,9 @@ "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, @@ -2213,9 +2866,9 @@ "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, @@ -2224,13 +2877,42 @@ "supports_prompt_caching": true, "supports_tool_choice": true }, + "azure/codex-mini": { + "max_tokens": 100000, + "max_input_tokens": 200000, + "max_output_tokens": 100000, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 6e-06, + "cache_read_input_token_cost": 3.75e-07, + "litellm_provider": "azure", + "mode": "responses", + "supports_pdf_input": true, + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "supports_vision": true, + "supports_prompt_caching": true, + "supports_system_messages": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supported_endpoints": [ + "/v1/responses" + ] + }, "azure/o1-preview": { "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, @@ -2243,11 +2925,12 @@ "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, "supports_function_calling": true, "supports_parallel_function_calling": true, "supports_vision": false, @@ -2258,9 +2941,9 @@ "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, @@ -2272,9 +2955,9 @@ "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, @@ -2286,11 +2969,11 @@ "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, @@ -2305,9 +2988,9 @@ "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, @@ -2321,9 +3004,9 @@ "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, @@ -2337,9 +3020,9 @@ "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, @@ -2353,9 +3036,9 @@ "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, @@ -2369,9 +3052,9 @@ "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, @@ -2385,9 +3068,9 @@ "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, @@ -2400,9 +3083,9 @@ "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, @@ -2415,8 +3098,8 @@ "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, @@ -2429,9 +3112,9 @@ "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, @@ -2439,15 +3122,16 @@ "supports_response_schema": true, "supports_vision": true, "supports_prompt_caching": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "deprecation_date": "2025-08-20" }, "azure/us/gpt-4o-2024-08-06": { "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, @@ -2461,9 +3145,9 @@ "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, @@ -2477,22 +3161,24 @@ "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": 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, "supports_parallel_function_calling": true, "supports_response_schema": true, "supports_vision": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "deprecation_date": "2025-12-20" }, "azure/global-standard/gpt-4o-mini": { "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, @@ -2505,9 +3191,9 @@ "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, @@ -2521,9 +3207,9 @@ "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, @@ -2537,9 +3223,9 @@ "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, @@ -2553,9 +3239,9 @@ "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, @@ -2569,8 +3255,8 @@ "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, @@ -2582,8 +3268,8 @@ "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, @@ -2594,8 +3280,8 @@ "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, @@ -2606,8 +3292,8 @@ "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, @@ -2617,7 +3303,7 @@ "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", @@ -2627,7 +3313,7 @@ "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", @@ -2637,8 +3323,8 @@ "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, @@ -2648,8 +3334,8 @@ "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, @@ -2660,8 +3346,8 @@ "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_vision": true, @@ -2671,8 +3357,8 @@ "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, @@ -2682,8 +3368,8 @@ "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, @@ -2695,8 +3381,8 @@ "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, @@ -2708,8 +3394,8 @@ "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, @@ -2721,8 +3407,8 @@ "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, @@ -2734,8 +3420,8 @@ "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, @@ -2747,8 +3433,8 @@ "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 @@ -2757,8 +3443,8 @@ "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, @@ -2768,8 +3454,8 @@ "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, @@ -2778,32 +3464,32 @@ "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 @@ -2811,8 +3497,8 @@ "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 @@ -2821,8 +3507,8 @@ "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 @@ -2830,67 +3516,157 @@ "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-08, + "output_cost_per_pixel": 0.0, + "litellm_provider": "azure", + "supported_endpoints": [ + "/v1/images/generations" + ] + }, + "azure/low/1024-x-1024/gpt-image-1": { + "mode": "image_generation", + "input_cost_per_pixel": 1.0490417e-08, + "output_cost_per_pixel": 0.0, + "litellm_provider": "azure", + "supported_endpoints": [ + "/v1/images/generations" + ] + }, + "azure/medium/1024-x-1024/gpt-image-1": { + "mode": "image_generation", + "input_cost_per_pixel": 4.0054321e-08, + "output_cost_per_pixel": 0.0, + "litellm_provider": "azure", + "supported_endpoints": [ + "/v1/images/generations" + ] + }, + "azure/high/1024-x-1024/gpt-image-1": { + "mode": "image_generation", + "input_cost_per_pixel": 1.59263611e-07, + "output_cost_per_pixel": 0.0, + "litellm_provider": "azure", + "supported_endpoints": [ + "/v1/images/generations" + ] + }, + "azure/low/1024-x-1536/gpt-image-1": { + "mode": "image_generation", + "input_cost_per_pixel": 1.0172526e-08, + "output_cost_per_pixel": 0.0, + "litellm_provider": "azure", + "supported_endpoints": [ + "/v1/images/generations" + ] + }, + "azure/medium/1024-x-1536/gpt-image-1": { + "mode": "image_generation", + "input_cost_per_pixel": 4.0054321e-08, + "output_cost_per_pixel": 0.0, + "litellm_provider": "azure", + "supported_endpoints": [ + "/v1/images/generations" + ] + }, + "azure/high/1024-x-1536/gpt-image-1": { + "mode": "image_generation", + "input_cost_per_pixel": 1.58945719e-07, + "output_cost_per_pixel": 0.0, + "litellm_provider": "azure", + "supported_endpoints": [ + "/v1/images/generations" + ] + }, + "azure/low/1536-x-1024/gpt-image-1": { + "mode": "image_generation", + "input_cost_per_pixel": 1.0172526e-08, + "output_cost_per_pixel": 0.0, + "litellm_provider": "azure", + "supported_endpoints": [ + "/v1/images/generations" + ] + }, + "azure/medium/1536-x-1024/gpt-image-1": { + "mode": "image_generation", + "input_cost_per_pixel": 4.0054321e-08, + "output_cost_per_pixel": 0.0, + "litellm_provider": "azure", + "supported_endpoints": [ + "/v1/images/generations" + ] + }, + "azure/high/1536-x-1024/gpt-image-1": { + "mode": "image_generation", + "input_cost_per_pixel": 1.58945719e-07, + "output_cost_per_pixel": 0.0, + "litellm_provider": "azure", + "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", "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", "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", "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", "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", "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", "mode": "image_generation" @@ -2905,8 +3681,8 @@ "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, @@ -2917,19 +3693,31 @@ "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, "source": "https://techcommunity.microsoft.com/blog/machinelearningblog/announcing-deepseek-v3-on-azure-ai-foundry-and-github/4390438" }, + "azure_ai/deepseek-v3-0324": { + "max_tokens": 8192, + "max_input_tokens": 128000, + "max_output_tokens": 8192, + "input_cost_per_token": 1.14e-06, + "output_cost_per_token": 4.56e-06, + "litellm_provider": "azure_ai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "source": "https://techcommunity.microsoft.com/blog/machinelearningblog/announcing-deepseek-v3-on-azure-ai-foundry-and-github/4390438" + }, "azure_ai/jamba-instruct": { "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 @@ -2938,19 +3726,31 @@ "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, "source": "https://azuremarketplace.microsoft.com/en/marketplace/apps/000-000.mistral-nemo-12b-2407?tab=PlansAndPrice" }, + "azure_ai/mistral-medium-2505": { + "max_tokens": 8191, + "max_input_tokens": 131072, + "max_output_tokens": 8191, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 2e-06, + "litellm_provider": "azure_ai", + "mode": "chat", + "supports_function_calling": true, + "supports_assistant_prefill": true, + "supports_tool_choice": true + }, "azure_ai/mistral-large": { "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, @@ -2960,8 +3760,8 @@ "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", @@ -2971,8 +3771,8 @@ "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, @@ -2983,8 +3783,8 @@ "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", @@ -2995,8 +3795,8 @@ "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", @@ -3007,8 +3807,8 @@ "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", @@ -3019,8 +3819,8 @@ "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, @@ -3032,20 +3832,46 @@ "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", "source": "https://azuremarketplace.microsoft.com/en/marketplace/apps/metagenai.llama-3-3-70b-instruct-offer?tab=Overview", "supports_tool_choice": true }, + "azure_ai/Llama-4-Scout-17B-16E-Instruct": { + "max_tokens": 16384, + "max_input_tokens": 10000000, + "max_output_tokens": 16384, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 7.8e-07, + "litellm_provider": "azure_ai", + "supports_function_calling": true, + "supports_vision": true, + "mode": "chat", + "source": "https://azure.microsoft.com/en-us/blog/introducing-the-llama-4-herd-in-azure-ai-foundry-and-azure-databricks/", + "supports_tool_choice": true + }, + "azure_ai/Llama-4-Maverick-17B-128E-Instruct-FP8": { + "max_tokens": 16384, + "max_input_tokens": 1000000, + "max_output_tokens": 16384, + "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, + "mode": "chat", + "source": "https://azure.microsoft.com/en-us/blog/introducing-the-llama-4-herd-in-azure-ai-foundry-and-azure-databricks/", + "supports_tool_choice": true + }, "azure_ai/Llama-3.2-90B-Vision-Instruct": { "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, @@ -3057,8 +3883,8 @@ "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 @@ -3067,41 +3893,41 @@ "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, @@ -3111,9 +3937,9 @@ "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, @@ -3125,8 +3951,8 @@ "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, @@ -3138,8 +3964,8 @@ "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, @@ -3150,8 +3976,8 @@ "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, @@ -3162,8 +3988,8 @@ "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, @@ -3174,8 +4000,8 @@ "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, @@ -3186,8 +4012,8 @@ "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, @@ -3198,8 +4024,8 @@ "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, @@ -3210,8 +4036,8 @@ "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, @@ -3222,8 +4048,8 @@ "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, @@ -3234,8 +4060,8 @@ "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, @@ -3268,30 +4094,48 @@ "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": 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" }, "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" }, @@ -3299,8 +4143,8 @@ "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" }, @@ -3308,8 +4152,8 @@ "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" }, @@ -3317,18 +4161,17 @@ "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" }, @@ -3336,244 +4179,278 @@ "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, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/mistral-small": { "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", "supports_assistant_prefill": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/mistral-small-latest": { "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", "supports_assistant_prefill": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/mistral-medium": { "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, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/mistral-medium-latest": { "max_tokens": 8191, - "max_input_tokens": 32000, + "max_input_tokens": 131072, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000027, - "output_cost_per_token": 0.0000081, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 2e-06, "litellm_provider": "mistral", "mode": "chat", + "supports_function_calling": true, "supports_assistant_prefill": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true + }, + "mistral/mistral-medium-2505": { + "max_tokens": 8191, + "max_input_tokens": 131072, + "max_output_tokens": 8191, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 2e-06, + "litellm_provider": "mistral", + "mode": "chat", + "supports_function_calling": true, + "supports_assistant_prefill": true, + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/mistral-medium-2312": { "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, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/mistral-large-latest": { "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, "supports_assistant_prefill": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/mistral-large-2411": { "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, "supports_assistant_prefill": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/mistral-large-2402": { "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, "supports_assistant_prefill": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/mistral-large-2407": { "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, "supports_assistant_prefill": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/pixtral-large-latest": { "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, "supports_assistant_prefill": true, "supports_vision": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/pixtral-large-2411": { "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, "supports_assistant_prefill": true, "supports_vision": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/pixtral-12b-2409": { "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, "supports_assistant_prefill": true, "supports_vision": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/open-mistral-7b": { "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, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/open-mixtral-8x7b": { "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, "supports_assistant_prefill": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/open-mixtral-8x22b": { "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, "supports_assistant_prefill": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/codestral-latest": { "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, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/codestral-2405": { "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, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/open-mistral-nemo": { "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/", "supports_assistant_prefill": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/open-mistral-nemo-2407": { "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/", "supports_assistant_prefill": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true }, "mistral/open-codestral-mamba": { "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/", @@ -3584,18 +4461,92 @@ "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/", "supports_assistant_prefill": true, "supports_tool_choice": true }, + "mistral/devstral-small-2505": { + "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 3e-07, + "litellm_provider": "mistral", + "mode": "chat", + "source": "https://mistral.ai/news/devstral", + "supports_function_calling": true, + "supports_assistant_prefill": true, + "supports_tool_choice": true, + "supports_response_schema": true + }, + "mistral/magistral-medium-latest": { + "max_tokens": 40000, + "max_input_tokens": 40000, + "max_output_tokens": 40000, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 5e-06, + "litellm_provider": "mistral", + "mode": "chat", + "source": "https://mistral.ai/news/magistral", + "supports_function_calling": true, + "supports_assistant_prefill": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_response_schema": true + }, + "mistral/magistral-medium-2506": { + "max_tokens": 40000, + "max_input_tokens": 40000, + "max_output_tokens": 40000, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 5e-06, + "litellm_provider": "mistral", + "mode": "chat", + "source": "https://mistral.ai/news/magistral", + "supports_function_calling": true, + "supports_assistant_prefill": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_response_schema": true + }, + "mistral/magistral-small-latest": { + "max_tokens": 40000, + "max_input_tokens": 40000, + "max_output_tokens": 40000, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, + "litellm_provider": "mistral", + "mode": "chat", + "source": "https://mistral.ai/pricing#api-pricing", + "supports_function_calling": true, + "supports_assistant_prefill": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_response_schema": true + }, + "mistral/magistral-small-2506": { + "max_tokens": 40000, + "max_input_tokens": 40000, + "max_output_tokens": 40000, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.5e-06, + "litellm_provider": "mistral", + "mode": "chat", + "source": "https://mistral.ai/pricing#api-pricing", + "supports_function_calling": true, + "supports_assistant_prefill": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_response_schema": true + }, "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" }, @@ -3603,9 +4554,9 @@ "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, @@ -3618,11 +4569,42 @@ "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_assistant_prefill": true, + "supports_tool_choice": true, + "supports_prompt_caching": true + }, + "deepseek/deepseek-r1": { + "max_tokens": 8192, + "max_input_tokens": 65536, + "max_output_tokens": 8192, + "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_assistant_prefill": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_prompt_caching": true + }, + "deepseek/deepseek-v3": { + "max_tokens": 8192, + "max_input_tokens": 65536, + "max_output_tokens": 8192, + "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": 1.1e-06, "litellm_provider": "deepseek", "mode": "chat", "supports_function_calling": true, @@ -3634,8 +4616,8 @@ "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/", @@ -3646,8 +4628,8 @@ "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/", @@ -3658,8 +4640,8 @@ "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/" @@ -3668,8 +4650,8 @@ "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/" @@ -3678,187 +4660,274 @@ "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, "supports_vision": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_web_search": true }, "xai/grok-2-vision-1212": { "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, "supports_vision": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_web_search": true }, "xai/grok-2-vision-latest": { "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, "supports_vision": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_web_search": true }, "xai/grok-2-vision": { "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, "supports_vision": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_web_search": true + }, + "xai/grok-3": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "litellm_provider": "xai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_response_schema": false, + "source": "https://x.ai/api#pricing", + "supports_web_search": true + }, + "xai/grok-3-latest": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "litellm_provider": "xai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_response_schema": false, + "source": "https://x.ai/api#pricing", + "supports_web_search": true }, "xai/grok-3-beta": { "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, "supports_tool_choice": true, "supports_response_schema": false, - "source": "https://x.ai/api#pricing" + "source": "https://x.ai/api#pricing", + "supports_web_search": true }, "xai/grok-3-fast-beta": { "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, "supports_tool_choice": true, "supports_response_schema": false, - "source": "https://x.ai/api#pricing" + "source": "https://x.ai/api#pricing", + "supports_web_search": true }, "xai/grok-3-fast-latest": { "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, "supports_tool_choice": true, "supports_response_schema": false, - "source": "https://x.ai/api#pricing" + "source": "https://x.ai/api#pricing", + "supports_web_search": true }, - "xai/grok-3-mini-beta": { + "xai/grok-3-mini": { "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, "supports_tool_choice": true, "supports_reasoning": true, "supports_response_schema": false, - "source": "https://x.ai/api#pricing" + "source": "https://x.ai/api#pricing", + "supports_web_search": true }, - "xai/grok-3-mini-fast-beta": { + "xai/grok-3-mini-latest": { "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": 3e-07, + "output_cost_per_token": 5e-07, "litellm_provider": "xai", "mode": "chat", "supports_function_calling": true, "supports_tool_choice": true, "supports_reasoning": true, "supports_response_schema": false, - "source": "https://x.ai/api#pricing" + "source": "https://x.ai/api#pricing", + "supports_web_search": true + }, + "xai/grok-3-mini-fast": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 4e-06, + "litellm_provider": "xai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_response_schema": false, + "source": "https://x.ai/api#pricing", + "supports_web_search": true }, "xai/grok-3-mini-fast-latest": { "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, "supports_function_calling": true, "supports_tool_choice": true, "supports_response_schema": false, - "source": "https://x.ai/api#pricing" + "source": "https://x.ai/api#pricing", + "supports_web_search": true + }, + "xai/grok-3-mini-beta": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 5e-07, + "litellm_provider": "xai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_response_schema": false, + "source": "https://x.ai/api#pricing", + "supports_web_search": true + }, + "xai/grok-3-mini-fast-beta": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 6e-07, + "output_cost_per_token": 4e-06, + "litellm_provider": "xai", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_response_schema": false, + "source": "https://x.ai/api#pricing", + "supports_web_search": true }, "xai/grok-vision-beta": { "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, "supports_vision": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_web_search": true }, "xai/grok-2-1212": { "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, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_web_search": true }, "xai/grok-2": { "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, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_web_search": true }, "xai/grok-2-latest": { "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, - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_web_search": true }, "deepseek/deepseek-coder": { "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, @@ -3867,25 +4936,24 @@ "supports_prompt_caching": true }, "groq/deepseek-r1-distill-llama-70b": { - "max_tokens": 131072, - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 0.00000075, - "output_cost_per_token": 0.00000099, + "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "input_cost_per_token": 7.5e-07, + "output_cost_per_token": 9.9e-07, "litellm_provider": "groq", "mode": "chat", - "supports_system_messages": false, - "supports_function_calling": false, + "supports_function_calling": true, + "supports_response_schema": true, "supports_reasoning": true, - "supports_response_schema": false, "supports_tool_choice": true }, "groq/llama-3.3-70b-versatile": { - "max_tokens": 8192, + "max_tokens": 32768, "max_input_tokens": 128000, - "max_output_tokens": 8192, - "input_cost_per_token": 0.00000059, - "output_cost_per_token": 0.00000079, + "max_output_tokens": 32768, + "input_cost_per_token": 5.9e-07, + "output_cost_per_token": 7.9e-07, "litellm_provider": "groq", "mode": "chat", "supports_function_calling": true, @@ -3896,18 +4964,28 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000059, - "output_cost_per_token": 0.00000099, + "input_cost_per_token": 5.9e-07, + "output_cost_per_token": 9.9e-07, "litellm_provider": "groq", "mode": "chat", - "supports_tool_choice": true + "supports_tool_choice": true, + "deprecation_date": "2025-04-14" + }, + "groq/llama-guard-3-8b": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "max_output_tokens": 8192, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, + "litellm_provider": "groq", + "mode": "chat" }, "groq/llama2-70b-4096": { "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, @@ -3918,106 +4996,109 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000005, - "output_cost_per_token": 0.00000008, + "input_cost_per_token": 5e-08, + "output_cost_per_token": 8e-08, "litellm_provider": "groq", "mode": "chat", - "supports_function_calling": true, - "supports_response_schema": true, "supports_tool_choice": true }, "groq/llama-3.2-1b-preview": { "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "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": "groq", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "deprecation_date": "2025-04-14" }, "groq/llama-3.2-3b-preview": { "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000006, - "output_cost_per_token": 0.00000006, + "input_cost_per_token": 6e-08, + "output_cost_per_token": 6e-08, "litellm_provider": "groq", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "deprecation_date": "2025-04-14" }, "groq/llama-3.2-11b-text-preview": { "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "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": "groq", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "deprecation_date": "2024-10-28" }, "groq/llama-3.2-11b-vision-preview": { "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "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": "groq", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, "supports_vision": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "deprecation_date": "2025-04-14" }, "groq/llama-3.2-90b-text-preview": { "max_tokens": 8192, "max_input_tokens": 8192, "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": "groq", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "deprecation_date": "2024-11-25" }, "groq/llama-3.2-90b-vision-preview": { "max_tokens": 8192, "max_input_tokens": 8192, "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": "groq", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, "supports_vision": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "deprecation_date": "2025-04-14" }, "groq/llama3-70b-8192": { "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, "supports_response_schema": true, "supports_tool_choice": true }, "groq/llama-3.1-8b-instant": { "max_tokens": 8192, - "max_input_tokens": 8192, + "max_input_tokens": 128000, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000005, - "output_cost_per_token": 0.00000008, + "input_cost_per_token": 5e-08, + "output_cost_per_token": 8e-08, "litellm_provider": "groq", "mode": "chat", "supports_function_calling": true, @@ -4028,110 +5109,169 @@ "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, "supports_response_schema": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "deprecation_date": "2025-01-24" }, "groq/llama-3.1-405b-reasoning": { "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, "supports_response_schema": true, "supports_tool_choice": true }, + "groq/meta-llama/llama-4-scout-17b-16e-instruct": { + "max_tokens": 8192, + "max_input_tokens": 131072, + "max_output_tokens": 8192, + "input_cost_per_token": 1.1e-07, + "output_cost_per_token": 3.4e-07, + "litellm_provider": "groq", + "mode": "chat", + "supports_function_calling": true, + "supports_response_schema": true, + "supports_tool_choice": true + }, + "groq/meta-llama/llama-4-maverick-17b-128e-instruct": { + "max_tokens": 8192, + "max_input_tokens": 131072, + "max_output_tokens": 8192, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 6e-07, + "litellm_provider": "groq", + "mode": "chat", + "supports_function_calling": true, + "supports_response_schema": true, + "supports_tool_choice": true + }, + "groq/mistral-saba-24b": { + "max_tokens": 32000, + "max_input_tokens": 32000, + "max_output_tokens": 32000, + "input_cost_per_token": 7.9e-07, + "output_cost_per_token": 7.9e-07, + "litellm_provider": "groq", + "mode": "chat" + }, "groq/mixtral-8x7b-32768": { "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 32768, - "input_cost_per_token": 0.00000024, - "output_cost_per_token": 0.00000024, + "input_cost_per_token": 2.4e-07, + "output_cost_per_token": 2.4e-07, "litellm_provider": "groq", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "deprecation_date": "2025-03-20" }, "groq/gemma-7b-it": { "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000007, - "output_cost_per_token": 0.00000007, + "input_cost_per_token": 7e-08, + "output_cost_per_token": 7e-08, "litellm_provider": "groq", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "deprecation_date": "2024-12-18" }, "groq/gemma2-9b-it": { "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "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": "groq", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": false, "supports_response_schema": true, - "supports_tool_choice": true + "supports_tool_choice": false }, "groq/llama3-groq-70b-8192-tool-use-preview": { "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000089, - "output_cost_per_token": 0.00000089, + "input_cost_per_token": 8.9e-07, + "output_cost_per_token": 8.9e-07, "litellm_provider": "groq", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, - "supports_tool_choice": true + "supports_tool_choice": true, + "deprecation_date": "2025-01-06" }, "groq/llama3-groq-8b-8192-tool-use-preview": { "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "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": "groq", "mode": "chat", "supports_function_calling": true, "supports_response_schema": true, + "supports_tool_choice": true, + "deprecation_date": "2025-01-06" + }, + "groq/qwen-qwq-32b": { + "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "input_cost_per_token": 2.9e-07, + "output_cost_per_token": 3.9e-07, + "litellm_provider": "groq", + "mode": "chat", + "supports_function_calling": true, + "supports_response_schema": true, + "supports_reasoning": true, "supports_tool_choice": true }, + "groq/playai-tts": { + "max_tokens": 10000, + "max_input_tokens": 10000, + "max_output_tokens": 10000, + "input_cost_per_character": 5e-05, + "litellm_provider": "groq", + "mode": "audio_speech" + }, "groq/whisper-large-v3": { - "mode": "audio_transcription", - "input_cost_per_second": 0.00003083, - "output_cost_per_second": 0, - "litellm_provider": "groq" + "input_cost_per_second": 3.083e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "groq", + "mode": "audio_transcription" }, "groq/whisper-large-v3-turbo": { - "mode": "audio_transcription", - "input_cost_per_second": 0.00001111, - "output_cost_per_second": 0, - "litellm_provider": "groq" + "input_cost_per_second": 1.111e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "groq", + "mode": "audio_transcription" }, "groq/distil-whisper-large-v3-en": { - "mode": "audio_transcription", - "input_cost_per_second": 0.00000556, - "output_cost_per_second": 0, - "litellm_provider": "groq" + "input_cost_per_second": 5.56e-06, + "output_cost_per_second": 0.0, + "litellm_provider": "groq", + "mode": "audio_transcription" }, "cerebras/llama3.1-8b": { "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, @@ -4141,30 +5281,42 @@ "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, "supports_tool_choice": true }, - "cerebras/llama3.3-70b": { + "cerebras/llama-3.3-70b": { "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, "supports_tool_choice": true }, + "cerebras/qwen-3-32b": { + "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 128000, + "input_cost_per_token": 4e-07, + "output_cost_per_token": 8e-07, + "litellm_provider": "cerebras", + "mode": "chat", + "supports_function_calling": true, + "supports_tool_choice": true, + "source": "https://inference-docs.cerebras.ai/support/pricing" + }, "friendliai/meta-llama-3.1-8b-instruct": { "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, @@ -4177,8 +5329,8 @@ "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, @@ -4191,8 +5343,8 @@ "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 @@ -4201,8 +5353,8 @@ "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" }, @@ -4210,8 +5362,8 @@ "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 @@ -4220,10 +5372,10 @@ "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, @@ -4239,10 +5391,15 @@ "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": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 + }, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4253,16 +5410,22 @@ "supports_prompt_caching": true, "supports_response_schema": true, "deprecation_date": "2025-10-01", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_web_search": true }, "claude-3-5-haiku-latest": { "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": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 + }, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4273,16 +5436,17 @@ "supports_prompt_caching": true, "supports_response_schema": true, "deprecation_date": "2025-10-01", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_web_search": true }, "claude-3-opus-latest": { "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, @@ -4298,10 +5462,10 @@ "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, @@ -4317,8 +5481,8 @@ "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, @@ -4331,13 +5495,19 @@ "supports_tool_choice": true }, "claude-3-5-sonnet-latest": { + "supports_computer_use": true, "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": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 + }, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4348,16 +5518,17 @@ "supports_prompt_caching": true, "supports_response_schema": true, "deprecation_date": "2025-06-01", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_web_search": true }, "claude-3-5-sonnet-20240620": { "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, @@ -4370,14 +5541,124 @@ "deprecation_date": "2025-06-01", "supports_tool_choice": true }, + "claude-opus-4-20250514": { + "max_tokens": 32000, + "max_input_tokens": 200000, + "max_output_tokens": 32000, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, + "search_context_cost_per_query": { + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 + }, + "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, + "supports_vision": true, + "tool_use_system_prompt_tokens": 159, + "supports_assistant_prefill": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_computer_use": true + }, + "claude-sonnet-4-20250514": { + "max_tokens": 64000, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "search_context_cost_per_query": { + "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-06, + "cache_read_input_token_cost": 3e-07, + "litellm_provider": "anthropic", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "tool_use_system_prompt_tokens": 159, + "supports_assistant_prefill": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_computer_use": true + }, + "claude-4-opus-20250514": { + "max_tokens": 32000, + "max_input_tokens": 200000, + "max_output_tokens": 32000, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, + "search_context_cost_per_query": { + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 + }, + "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, + "supports_vision": true, + "tool_use_system_prompt_tokens": 159, + "supports_assistant_prefill": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_computer_use": true + }, + "claude-4-sonnet-20250514": { + "max_tokens": 64000, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "search_context_cost_per_query": { + "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-06, + "cache_read_input_token_cost": 3e-07, + "litellm_provider": "anthropic", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "tool_use_system_prompt_tokens": 159, + "supports_assistant_prefill": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_computer_use": true + }, "claude-3-7-sonnet-latest": { + "supports_computer_use": true, "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, + "search_context_cost_per_query": { + "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-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4392,13 +5673,19 @@ "supports_reasoning": true }, "claude-3-7-sonnet-20250219": { + "supports_computer_use": true, "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": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 + }, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4410,16 +5697,23 @@ "supports_response_schema": true, "deprecation_date": "2026-02-01", "supports_tool_choice": true, - "supports_reasoning": true + "supports_reasoning": true, + "supports_web_search": true }, "claude-3-5-sonnet-20241022": { + "supports_computer_use": true, "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": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 + }, "litellm_provider": "anthropic", "mode": "chat", "supports_function_calling": true, @@ -4430,14 +5724,15 @@ "supports_prompt_caching": true, "supports_response_schema": true, "deprecation_date": "2025-10-01", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_web_search": true }, "text-bison": { "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" @@ -4446,8 +5741,8 @@ "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" @@ -4456,8 +5751,8 @@ "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" @@ -4466,10 +5761,10 @@ "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" @@ -4478,10 +5773,10 @@ "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" @@ -4490,8 +5785,8 @@ "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" @@ -4500,8 +5795,8 @@ "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" @@ -4510,10 +5805,10 @@ "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", @@ -4523,10 +5818,10 @@ "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", @@ -4536,10 +5831,10 @@ "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", @@ -4550,10 +5845,10 @@ "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", @@ -4563,10 +5858,10 @@ "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", @@ -4576,10 +5871,10 @@ "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", @@ -4589,10 +5884,10 @@ "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" @@ -4601,10 +5896,10 @@ "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" @@ -4613,10 +5908,10 @@ "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" @@ -4625,10 +5920,10 @@ "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" @@ -4637,8 +5932,8 @@ "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" @@ -4647,8 +5942,8 @@ "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" @@ -4657,8 +5952,8 @@ "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" @@ -4667,8 +5962,8 @@ "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" @@ -4677,10 +5972,10 @@ "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", @@ -4690,10 +5985,10 @@ "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", @@ -4703,10 +5998,10 @@ "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", @@ -4716,10 +6011,10 @@ "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", @@ -4729,10 +6024,10 @@ "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", @@ -4742,28 +6037,95 @@ "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", "supports_tool_choice": true }, + "meta_llama/Llama-4-Scout-17B-16E-Instruct-FP8": { + "max_tokens": 128000, + "max_input_tokens": 10000000, + "max_output_tokens": 4028, + "litellm_provider": "meta_llama", + "mode": "chat", + "supports_function_calling": true, + "source": "https://llama.developer.meta.com/docs/models", + "supports_tool_choice": true, + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ] + }, + "meta_llama/Llama-4-Maverick-17B-128E-Instruct-FP8": { + "max_tokens": 128000, + "max_input_tokens": 1000000, + "max_output_tokens": 4028, + "litellm_provider": "meta_llama", + "mode": "chat", + "supports_function_calling": true, + "source": "https://llama.developer.meta.com/docs/models", + "supports_tool_choice": true, + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ] + }, + "meta_llama/Llama-3.3-70B-Instruct": { + "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 4028, + "litellm_provider": "meta_llama", + "mode": "chat", + "supports_function_calling": true, + "source": "https://llama.developer.meta.com/docs/models", + "supports_tool_choice": true, + "supported_modalities": [ + "text" + ], + "supported_output_modalities": [ + "text" + ] + }, + "meta_llama/Llama-3.3-8B-Instruct": { + "max_tokens": 128000, + "max_input_tokens": 128000, + "max_output_tokens": 4028, + "litellm_provider": "meta_llama", + "mode": "chat", + "supports_function_calling": true, + "source": "https://llama.developer.meta.com/docs/models", + "supports_tool_choice": true, + "supported_modalities": [ + "text" + ], + "supported_output_modalities": [ + "text" + ] + }, "gemini-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", "supports_tool_choice": true }, @@ -4773,13 +6135,14 @@ "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 }, @@ -4789,16 +6152,17 @@ "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, @@ -4806,15 +6170,16 @@ "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, @@ -4822,15 +6187,16 @@ "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-pro-002": { "max_tokens": 8192, @@ -4838,35 +6204,36 @@ "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": { "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_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": 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_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, @@ -4875,26 +6242,27 @@ "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" + "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_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": 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_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, @@ -4903,26 +6271,27 @@ "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": { "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_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": 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_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, @@ -4931,84 +6300,88 @@ "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": { "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_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": 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_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" + "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": { "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_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": 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_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" + "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_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": 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_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" + "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, @@ -5020,20 +6393,20 @@ "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, @@ -5041,7 +6414,8 @@ "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, @@ -5053,20 +6427,20 @@ "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, @@ -5074,7 +6448,8 @@ "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, @@ -5086,20 +6461,20 @@ "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, @@ -5108,7 +6483,8 @@ "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, @@ -5120,20 +6496,20 @@ "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, @@ -5142,7 +6518,8 @@ "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, @@ -5154,27 +6531,28 @@ "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, @@ -5188,7 +6566,8 @@ "mode": "chat", "supports_function_calling": false, "supports_tool_choice": true, - "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/gemini-experimental" + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/gemini-experimental", + "supports_parallel_function_calling": true }, "gemini-flash-experimental": { "max_tokens": 8192, @@ -5202,7 +6581,8 @@ "mode": "chat", "supports_function_calling": false, "supports_tool_choice": true, - "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/gemini-experimental" + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/gemini-experimental", + "supports_parallel_function_calling": true }, "gemini-pro-vision": { "max_tokens": 2048, @@ -5211,15 +6591,16 @@ "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, @@ -5228,15 +6609,16 @@ "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, @@ -5245,8 +6627,8 @@ "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", @@ -5254,14 +6636,15 @@ "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", @@ -5271,27 +6654,27 @@ "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", "supports_tool_choice": true }, "gemini-2.5-pro-exp-03-25": { - "max_tokens": 65536, + "max_tokens": 65535, "max_input_tokens": 1048576, - "max_output_tokens": 65536, + "max_output_tokens": 65535, "max_images_per_prompt": 3000, "max_videos_per_prompt": 10, "max_video_length": 1, "max_audio_length_hours": 8.4, "max_audio_per_prompt": 1, "max_pdf_size_mb": 30, - "input_cost_per_token": 0.00000125, - "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, @@ -5302,10 +6685,22 @@ "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, + "supports_web_search": true }, "gemini-2.0-pro-exp-02-05": { "max_tokens": 8192, @@ -5317,10 +6712,10 @@ "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, @@ -5331,10 +6726,22 @@ "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, + "supports_web_search": true }, "gemini-2.0-flash-exp": { "max_tokens": 8192, @@ -5349,14 +6756,14 @@ "input_cost_per_image": 0, "input_cost_per_video_per_second": 0, "input_cost_per_audio_per_second": 0, - "input_cost_per_token": 0, + "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, + "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, @@ -5367,10 +6774,20 @@ "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, + "supports_web_search": true }, "gemini-2.0-flash-001": { "max_tokens": 8192, @@ -5382,9 +6799,9 @@ "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, @@ -5393,9 +6810,20 @@ "supports_response_schema": true, "supports_audio_output": true, "supports_tool_choice": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], - "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing" + "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", + "supports_parallel_function_calling": true, + "supports_web_search": true }, "gemini-2.0-flash-thinking-exp": { "max_tokens": 8192, @@ -5428,10 +6856,20 @@ "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, + "supports_web_search": true }, "gemini-2.0-flash-thinking-exp-01-21": { "max_tokens": 65536, @@ -5464,25 +6902,249 @@ "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, + "supports_web_search": true }, - "gemini/gemini-2.5-flash-preview-04-17": { - "max_tokens": 65536, + "gemini-2.5-pro": { + "max_tokens": 65535, "max_input_tokens": 1048576, - "max_output_tokens": 65536, + "max_output_tokens": 65535, "max_images_per_prompt": 3000, "max_videos_per_prompt": 10, "max_video_length": 1, "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_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, + "supports_function_calling": true, + "supports_vision": true, + "supports_audio_input": true, + "supports_video_input": true, + "supports_pdf_input": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": 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", + "supports_web_search": true + }, + "gemini/gemini-2.5-pro-exp-03-25": { + "max_tokens": 65535, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "input_cost_per_token": 0.0, + "input_cost_per_token_above_200k_tokens": 0.0, + "output_cost_per_token": 0.0, + "output_cost_per_token_above_200k_tokens": 0.0, + "litellm_provider": "gemini", + "mode": "chat", + "rpm": 5, + "tpm": 250000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": true, + "supports_audio_input": true, + "supports_video_input": true, + "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", + "supports_web_search": true + }, + "gemini/gemini-2.5-pro": { + "max_tokens": 65535, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "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": 2000, + "tpm": 800000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": true, + "supports_audio_input": true, + "supports_video_input": true, + "supports_pdf_input": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": 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", + "supports_web_search": true + }, + "gemini/gemini-2.5-flash": { + "max_tokens": 65535, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 2.5e-06, + "output_cost_per_reasoning_token": 2.5e-06, + "litellm_provider": "gemini", + "mode": "chat", + "supports_reasoning": true, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": true, + "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", + "supports_parallel_function_calling": true, + "supports_web_search": true, + "supports_url_context": true, + "tpm": 8000000, + "rpm": 100000, + "supports_pdf_input": true + }, + "gemini-2.5-flash": { + "max_tokens": 65535, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 2.5e-06, + "output_cost_per_reasoning_token": 2.5e-06, + "litellm_provider": "vertex_ai-language-models", + "mode": "chat", + "supports_reasoning": true, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": true, + "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", + "supports_parallel_function_calling": true, + "supports_web_search": true, + "supports_url_context": true, + "supports_pdf_input": true + }, + "gemini/gemini-2.5-flash-preview-tts": { + "max_tokens": 65535, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "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, @@ -5494,25 +7156,163 @@ "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"], - "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview" + "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", + "supports_web_search": true }, - "gemini-2.5-flash-preview-04-17": { - "max_tokens": 65536, + "gemini/gemini-2.5-flash-preview-05-20": { + "max_tokens": 65535, "max_input_tokens": 1048576, - "max_output_tokens": 65536, + "max_output_tokens": 65535, "max_images_per_prompt": 3000, "max_videos_per_prompt": 10, "max_video_length": 1, "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": 3e-07, + "output_cost_per_token": 2.5e-06, + "output_cost_per_reasoning_token": 2.5e-06, + "litellm_provider": "gemini", + "mode": "chat", + "rpm": 10, + "tpm": 250000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": true, + "supports_reasoning": true, + "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" + ], + "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview", + "supports_web_search": true, + "supports_url_context": true, + "supports_pdf_input": true + }, + "gemini/gemini-2.5-flash-preview-04-17": { + "max_tokens": 65535, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "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, + "tpm": 250000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": true, + "supports_reasoning": true, + "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" + ], + "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview", + "supports_web_search": true, + "supports_pdf_input": true + }, + "gemini/gemini-2.5-flash-lite-preview-06-17": { + "max_tokens": 65535, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "input_cost_per_audio_token": 5e-07, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, + "output_cost_per_reasoning_token": 4e-07, + "litellm_provider": "gemini", + "mode": "chat", + "rpm": 15, + "tpm": 250000, + "supports_reasoning": true, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": true, + "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-lite", + "supports_parallel_function_calling": true, + "supports_web_search": true, + "supports_url_context": true, + "supports_pdf_input": true + }, + "gemini-2.5-flash-preview-05-20": { + "max_tokens": 65535, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "input_cost_per_audio_token": 1e-06, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 2.5e-06, + "output_cost_per_reasoning_token": 2.5e-06, "litellm_provider": "vertex_ai-language-models", "mode": "chat", "supports_reasoning": true, @@ -5522,10 +7322,110 @@ "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, + "supports_web_search": true, + "supports_url_context": true, + "supports_pdf_input": true + }, + "gemini-2.5-flash-preview-04-17": { + "max_tokens": 65535, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "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, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": true, + "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", + "supports_parallel_function_calling": true, + "supports_web_search": true, + "supports_pdf_input": true + }, + "gemini-2.5-flash-lite-preview-06-17": { + "max_tokens": 65535, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "input_cost_per_audio_token": 5e-07, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 4e-07, + "output_cost_per_reasoning_token": 4e-07, + "litellm_provider": "vertex_ai-language-models", + "mode": "chat", + "supports_reasoning": true, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": true, + "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", + "supports_parallel_function_calling": true, + "supports_web_search": true, + "supports_url_context": true, + "supports_pdf_input": true }, "gemini-2.0-flash": { "max_tokens": 8192, @@ -5537,9 +7437,9 @@ "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, @@ -5548,10 +7448,21 @@ "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, + "supports_web_search": true, + "supports_url_context": true }, "gemini-2.0-flash-lite": { "max_input_tokens": 1048576, @@ -5562,9 +7473,9 @@ "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, @@ -5572,10 +7483,19 @@ "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 + "supports_tool_choice": true, + "supports_parallel_function_calling": true, + "supports_web_search": true }, "gemini-2.0-flash-lite-001": { "max_input_tokens": 1048576, @@ -5586,9 +7506,9 @@ "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, @@ -5596,26 +7516,36 @@ "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, + "deprecation_date": "2026-02-25", + "supports_parallel_function_calling": true, + "supports_web_search": true }, - "gemini-2.5-pro-preview-03-25": { - "max_tokens": 65536, + "gemini-2.5-pro-preview-06-05": { + "max_tokens": 65535, "max_input_tokens": 1048576, - "max_output_tokens": 65536, + "max_output_tokens": 65535, "max_images_per_prompt": 3000, "max_videos_per_prompt": 10, "max_video_length": 1, "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, @@ -5625,10 +7555,182 @@ "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, + "supports_web_search": true, + "supports_pdf_input": true + }, + "gemini-2.5-pro-preview-05-06": { + "max_tokens": 65535, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "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, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": true, + "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" + ], + "supported_regions": [ + "global" + ], + "source": "https://ai.google.dev/gemini-api/docs/models#gemini-2.5-flash-preview", + "supports_parallel_function_calling": true, + "supports_web_search": true, + "supports_pdf_input": true + }, + "gemini-2.5-pro-preview-03-25": { + "max_tokens": 65535, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "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, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": true, + "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", + "supports_parallel_function_calling": true, + "supports_web_search": true, + "supports_pdf_input": true + }, + "gemini-2.0-flash-preview-image-generation": { + "max_tokens": 8192, + "max_input_tokens": 1048576, + "max_output_tokens": 8192, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "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, + "supports_function_calling": true, + "supports_vision": true, + "supports_response_schema": true, + "supports_audio_output": true, + "supports_audio_input": true, + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], + "supports_tool_choice": true, + "source": "https://ai.google.dev/pricing#2_0flash", + "supports_parallel_function_calling": true, + "supports_web_search": true + }, + "gemini-2.5-pro-preview-tts": { + "max_tokens": 65535, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "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, + "supports_function_calling": true, + "supports_vision": true, + "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", + "supports_parallel_function_calling": true, + "supports_web_search": true }, "gemini/gemini-2.0-pro-exp-02-05": { "max_tokens": 8192, @@ -5666,7 +7768,45 @@ "supports_pdf_input": true, "supports_response_schema": true, "supports_tool_choice": true, - "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing" + "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing", + "supports_web_search": true + }, + "gemini/gemini-2.0-flash-preview-image-generation": { + "max_tokens": 8192, + "max_input_tokens": 1048576, + "max_output_tokens": 8192, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "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, + "tpm": 10000000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": true, + "supports_response_schema": true, + "supports_audio_output": true, + "supports_audio_input": true, + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], + "supports_tool_choice": true, + "source": "https://ai.google.dev/pricing#2_0flash", + "supports_web_search": true }, "gemini/gemini-2.0-flash": { "max_tokens": 8192, @@ -5678,9 +7818,9 @@ "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, @@ -5691,10 +7831,20 @@ "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_web_search": true, + "supports_url_context": true }, "gemini/gemini-2.0-flash-lite": { "max_input_tokens": 1048576, @@ -5705,9 +7855,9 @@ "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, @@ -5718,9 +7868,17 @@ "supports_response_schema": true, "supports_audio_output": true, "supports_tool_choice": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], - "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.0-flash-lite" + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.0-flash-lite", + "supports_web_search": true }, "gemini/gemini-2.0-flash-001": { "max_tokens": 8192, @@ -5732,9 +7890,9 @@ "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, @@ -5745,25 +7903,34 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text", "image"], - "source": "https://ai.google.dev/pricing#2_0flash" + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text", + "image" + ], + "source": "https://ai.google.dev/pricing#2_0flash", + "supports_web_search": true }, - "gemini/gemini-2.5-pro-preview-03-25": { - "max_tokens": 65536, + "gemini/gemini-2.5-pro-preview-tts": { + "max_tokens": 65535, "max_input_tokens": 1048576, - "max_output_tokens": 65536, + "max_output_tokens": 65535, "max_images_per_prompt": 3000, "max_videos_per_prompt": 10, "max_video_length": 1, "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, @@ -5774,9 +7941,130 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "supported_modalities": ["text", "image", "audio", "video"], - "supported_output_modalities": ["text"], - "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_web_search": true + }, + "gemini/gemini-2.5-pro-preview-06-05": { + "max_tokens": 65535, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "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, + "tpm": 10000000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": true, + "supports_response_schema": true, + "supports_audio_output": false, + "supports_tool_choice": true, + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.5-pro-preview", + "supports_web_search": true, + "supports_url_context": true, + "supports_pdf_input": true + }, + "gemini/gemini-2.5-pro-preview-05-06": { + "max_tokens": 65535, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "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, + "tpm": 10000000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": true, + "supports_response_schema": true, + "supports_audio_output": false, + "supports_tool_choice": true, + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.5-pro-preview", + "supports_web_search": true, + "supports_url_context": true, + "supports_pdf_input": true + }, + "gemini/gemini-2.5-pro-preview-03-25": { + "max_tokens": 65535, + "max_input_tokens": 1048576, + "max_output_tokens": 65535, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "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, + "tpm": 10000000, + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": true, + "supports_response_schema": true, + "supports_audio_output": false, + "supports_tool_choice": true, + "supported_modalities": [ + "text", + "image", + "audio", + "video" + ], + "supported_output_modalities": [ + "text" + ], + "source": "https://ai.google.dev/gemini-api/docs/pricing#gemini-2.5-pro-preview", + "supports_web_search": true, + "supports_pdf_input": true }, "gemini/gemini-2.0-flash-exp": { "max_tokens": 8192, @@ -5811,10 +8099,19 @@ "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 + "supports_tool_choice": true, + "supports_web_search": true }, "gemini/gemini-2.0-flash-lite-preview-02-05": { "max_tokens": 8192, @@ -5826,9 +8123,9 @@ "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, @@ -5839,9 +8136,17 @@ "supports_response_schema": true, "supports_audio_output": false, "supports_tool_choice": true, - "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" + "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", + "supports_web_search": true }, "gemini/gemini-2.0-flash-thinking-exp": { "max_tokens": 8192, @@ -5876,10 +8181,19 @@ "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 + "supports_tool_choice": true, + "supports_web_search": true }, "gemini/gemini-2.0-flash-thinking-exp-01-21": { "max_tokens": 8192, @@ -5914,10 +8228,19 @@ "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 + "supports_tool_choice": true, + "supports_web_search": true }, "gemini/gemma-3-27b-it": { "max_tokens": 8192, @@ -5979,8 +8302,8 @@ "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, @@ -5992,8 +8315,8 @@ "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, @@ -6002,11 +8325,12 @@ "supports_tool_choice": true }, "vertex_ai/claude-3-5-sonnet": { + "supports_computer_use": true, "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, @@ -6019,8 +8343,8 @@ "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, @@ -6030,11 +8354,12 @@ "supports_tool_choice": true }, "vertex_ai/claude-3-5-sonnet-v2": { + "supports_computer_use": true, "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, @@ -6044,11 +8369,12 @@ "supports_tool_choice": true }, "vertex_ai/claude-3-5-sonnet-v2@20241022": { + "supports_computer_use": true, "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, @@ -6058,13 +8384,14 @@ "supports_tool_choice": true }, "vertex_ai/claude-3-7-sonnet@20250219": { + "supports_computer_use": true, "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, @@ -6078,12 +8405,116 @@ "supports_reasoning": true, "supports_tool_choice": true }, + "vertex_ai/claude-opus-4": { + "max_tokens": 32000, + "max_input_tokens": 200000, + "max_output_tokens": 32000, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, + "search_context_cost_per_query": { + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 + }, + "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, + "supports_vision": true, + "tool_use_system_prompt_tokens": 159, + "supports_assistant_prefill": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_computer_use": true + }, + "vertex_ai/claude-opus-4@20250514": { + "max_tokens": 32000, + "max_input_tokens": 200000, + "max_output_tokens": 32000, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, + "search_context_cost_per_query": { + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 + }, + "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, + "supports_vision": true, + "tool_use_system_prompt_tokens": 159, + "supports_assistant_prefill": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_computer_use": true + }, + "vertex_ai/claude-sonnet-4": { + "max_tokens": 64000, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "search_context_cost_per_query": { + "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-06, + "cache_read_input_token_cost": 3e-07, + "litellm_provider": "vertex_ai-anthropic_models", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "tool_use_system_prompt_tokens": 159, + "supports_assistant_prefill": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_computer_use": true + }, + "vertex_ai/claude-sonnet-4@20250514": { + "max_tokens": 64000, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "search_context_cost_per_query": { + "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-06, + "cache_read_input_token_cost": 3e-07, + "litellm_provider": "vertex_ai-anthropic_models", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "tool_use_system_prompt_tokens": 159, + "supports_assistant_prefill": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_computer_use": true + }, "vertex_ai/claude-3-haiku": { "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, @@ -6095,8 +8526,8 @@ "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": "vertex_ai-anthropic_models", "mode": "chat", "supports_function_calling": true, @@ -6108,8 +8539,8 @@ "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, @@ -6121,8 +8552,8 @@ "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, @@ -6134,8 +8565,8 @@ "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, @@ -6147,8 +8578,8 @@ "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, @@ -6167,6 +8598,86 @@ "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#partner-models", "supports_tool_choice": true }, + "vertex_ai/meta/llama-4-scout-17b-16e-instruct-maas": { + "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" + ] + }, + "vertex_ai/meta/llama-4-scout-17b-128e-instruct-maas": { + "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" + ] + }, + "vertex_ai/meta/llama-4-maverick-17b-128e-instruct-maas": { + "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" + ] + }, + "vertex_ai/meta/llama-4-maverick-17b-16e-instruct-maas": { + "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" + ] + }, "vertex_ai/meta/llama3-70b-instruct-maas": { "max_tokens": 32000, "max_input_tokens": 32000, @@ -6206,8 +8717,8 @@ "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, @@ -6217,8 +8728,8 @@ "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, @@ -6228,8 +8739,8 @@ "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, @@ -6239,8 +8750,8 @@ "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, @@ -6250,8 +8761,8 @@ "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, @@ -6261,8 +8772,8 @@ "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", @@ -6272,8 +8783,8 @@ "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, @@ -6284,8 +8795,8 @@ "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 @@ -6294,8 +8805,8 @@ "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 @@ -6304,8 +8815,8 @@ "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 @@ -6314,8 +8825,8 @@ "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 @@ -6324,8 +8835,8 @@ "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 @@ -6334,8 +8845,8 @@ "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, @@ -6345,8 +8856,8 @@ "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, @@ -6356,8 +8867,8 @@ "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, @@ -6367,15 +8878,33 @@ "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" + }, + "vertex_ai/imagen-4.0-generate-preview-06-06": { + "output_cost_per_image": 0.04, + "litellm_provider": "vertex_ai-image-models", + "mode": "image_generation", + "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing" + }, + "vertex_ai/imagen-4.0-ultra-generate-preview-06-06": { + "output_cost_per_image": 0.06, + "litellm_provider": "vertex_ai-image-models", + "mode": "image_generation", + "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing" + }, + "vertex_ai/imagen-4.0-fast-generate-preview-06-06": { + "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" @@ -6402,8 +8931,18 @@ "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", + "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models" + }, + "gemini-embedding-001": { + "max_tokens": 2048, + "max_input_tokens": 2048, + "output_vector_size": 3072, + "input_cost_per_token": 1.5e-07, "output_cost_per_token": 0, "litellm_provider": "vertex_ai-embedding-models", "mode": "embedding", @@ -6413,8 +8952,8 @@ "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", @@ -6424,8 +8963,8 @@ "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", @@ -6435,42 +8974,54 @@ "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", @@ -6480,8 +9031,8 @@ "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", @@ -6491,8 +9042,8 @@ "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", @@ -6502,8 +9053,8 @@ "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", @@ -6513,8 +9064,8 @@ "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", @@ -6524,8 +9075,8 @@ "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", @@ -6535,18 +9086,18 @@ "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", @@ -6556,8 +9107,8 @@ "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" @@ -6566,8 +9117,8 @@ "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" @@ -6576,8 +9127,8 @@ "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" @@ -6586,8 +9137,8 @@ "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" @@ -6596,8 +9147,8 @@ "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" @@ -6606,8 +9157,8 @@ "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" @@ -6622,12 +9173,12 @@ "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, + "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, @@ -6651,12 +9202,12 @@ "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, + "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, @@ -6680,10 +9231,10 @@ "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, + "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, @@ -6705,10 +9256,10 @@ "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, + "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, @@ -6885,10 +9436,10 @@ "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, @@ -6902,10 +9453,10 @@ "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, @@ -6921,10 +9472,10 @@ "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, @@ -6942,10 +9493,10 @@ "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, @@ -6963,10 +9514,10 @@ "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, @@ -7001,10 +9552,10 @@ "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, @@ -7020,10 +9571,10 @@ "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, @@ -7037,8 +9588,8 @@ "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, @@ -7049,8 +9600,8 @@ "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, @@ -7058,12 +9609,23 @@ "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models#foundation_models", "supports_tool_choice": true }, + "command-a-03-2025": { + "max_tokens": 8000, + "max_input_tokens": 256000, + "max_output_tokens": 8000, + "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-r": { "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, @@ -7073,8 +9635,8 @@ "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, @@ -7084,8 +9646,8 @@ "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, @@ -7096,8 +9658,8 @@ "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 @@ -7106,8 +9668,8 @@ "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, @@ -7117,8 +9679,8 @@ "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, @@ -7128,17 +9690,17 @@ "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": { + "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" }, @@ -7200,16 +9762,16 @@ "embed-english-light-v3.0": { "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_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" @@ -7217,33 +9779,33 @@ "embed-english-v2.0": { "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_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_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_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, @@ -7256,8 +9818,8 @@ "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 @@ -7266,8 +9828,8 @@ "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 @@ -7276,8 +9838,8 @@ "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 @@ -7286,8 +9848,8 @@ "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 @@ -7296,8 +9858,8 @@ "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 @@ -7306,8 +9868,8 @@ "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 @@ -7316,8 +9878,8 @@ "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 @@ -7326,8 +9888,8 @@ "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 @@ -7336,8 +9898,8 @@ "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 @@ -7346,8 +9908,8 @@ "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 @@ -7356,8 +9918,8 @@ "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 @@ -7366,8 +9928,8 @@ "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 @@ -7376,19 +9938,34 @@ "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 }, + "openrouter/deepseek/deepseek-r1-0528": { + "max_tokens": 8192, + "max_input_tokens": 65336, + "max_output_tokens": 8192, + "input_cost_per_token": 5e-07, + "input_cost_per_token_cache_hit": 1.4e-07, + "output_cost_per_token": 2.15e-06, + "litellm_provider": "openrouter", + "mode": "chat", + "supports_function_calling": true, + "supports_assistant_prefill": true, + "supports_reasoning": true, + "supports_tool_choice": true, + "supports_prompt_caching": true + }, "openrouter/deepseek/deepseek-r1": { "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, @@ -7401,8 +9978,8 @@ "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", @@ -7412,8 +9989,8 @@ "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", @@ -7421,18 +9998,40 @@ }, "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 }, + "openrouter/google/gemini-2.5-pro": { + "max_tokens": 8192, + "max_input_tokens": 1048576, + "max_output_tokens": 8192, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 1.25e-06, + "output_cost_per_token": 1e-05, + "litellm_provider": "openrouter", + "mode": "chat", + "supports_system_messages": true, + "supports_function_calling": true, + "supports_vision": true, + "supports_response_schema": true, + "supports_audio_output": true, + "supports_tool_choice": true + }, "openrouter/google/gemini-pro-1.5": { "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_token": 2.5e-06, + "output_cost_per_token": 7.5e-06, "input_cost_per_image": 0.00265, "litellm_provider": "openrouter", "mode": "chat", @@ -7450,9 +10049,31 @@ "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, + "supports_function_calling": true, + "supports_vision": true, + "supports_response_schema": true, + "supports_audio_output": true, + "supports_tool_choice": true + }, + "openrouter/google/gemini-2.5-flash": { + "max_tokens": 8192, + "max_input_tokens": 1048576, + "max_output_tokens": 8192, + "max_images_per_prompt": 3000, + "max_videos_per_prompt": 10, + "max_video_length": 1, + "max_audio_length_hours": 8.4, + "max_audio_per_prompt": 1, + "max_pdf_size_mb": 30, + "input_cost_per_audio_token": 7e-07, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 2.5e-06, "litellm_provider": "openrouter", "mode": "chat", "supports_system_messages": true, @@ -7464,32 +10085,32 @@ }, "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_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, "input_cost_per_image": 0.0004, "litellm_provider": "openrouter", "mode": "chat", @@ -7499,8 +10120,8 @@ }, "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, @@ -7510,8 +10131,8 @@ "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, @@ -7523,8 +10144,8 @@ "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, @@ -7532,11 +10153,12 @@ "supports_tool_choice": true }, "openrouter/anthropic/claude-3.5-sonnet": { + "supports_computer_use": true, "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, @@ -7546,11 +10168,12 @@ "supports_tool_choice": true }, "openrouter/anthropic/claude-3.5-sonnet:beta": { + "supports_computer_use": true, "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, @@ -7559,11 +10182,12 @@ "supports_tool_choice": true }, "openrouter/anthropic/claude-3.7-sonnet": { + "supports_computer_use": true, "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", @@ -7575,11 +10199,12 @@ "supports_tool_choice": true }, "openrouter/anthropic/claude-3.7-sonnet:beta": { + "supports_computer_use": true, "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", @@ -7591,8 +10216,8 @@ }, "openrouter/anthropic/claude-3-sonnet": { "max_tokens": 200000, - "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", @@ -7600,34 +10225,59 @@ "supports_vision": true, "supports_tool_choice": true }, + "openrouter/anthropic/claude-sonnet-4": { + "supports_computer_use": true, + "max_tokens": 8192, + "max_input_tokens": 200000, + "max_output_tokens": 8192, + "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, + "supports_vision": true, + "supports_reasoning": true, + "tool_use_system_prompt_tokens": 159, + "supports_assistant_prefill": true, + "supports_tool_choice": true + }, "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": { + "openrouter/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/mistralai/mistral-small-3.2-24b-instruct": { + "max_tokens": 32000, + "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_token": 1.25e-07, + "output_cost_per_token": 3.75e-07, "input_cost_per_image": 0.0025, "litellm_provider": "openrouter", "mode": "chat", @@ -7637,8 +10287,8 @@ }, "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 @@ -7653,24 +10303,24 @@ }, "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 @@ -7679,9 +10329,9 @@ "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, @@ -7696,8 +10346,8 @@ "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, @@ -7709,8 +10359,8 @@ "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, @@ -7722,8 +10372,8 @@ "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, @@ -7735,8 +10385,8 @@ "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, @@ -7748,8 +10398,8 @@ "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, @@ -7762,8 +10412,8 @@ "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, @@ -7776,8 +10426,8 @@ "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, @@ -7789,8 +10439,8 @@ "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, @@ -7800,8 +10450,8 @@ }, "openrouter/openai/gpt-4-vision-preview": { "max_tokens": 130000, - "input_cost_per_token": 0.00001, - "output_cost_per_token": 0.00003, + "input_cost_per_token": 1e-05, + "output_cost_per_token": 3e-05, "input_cost_per_image": 0.01445, "litellm_provider": "openrouter", "mode": "chat", @@ -7811,24 +10461,24 @@ }, "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 @@ -7836,8 +10486,8 @@ "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 @@ -7845,8 +10495,8 @@ "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 @@ -7855,8 +10505,8 @@ "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, @@ -7866,96 +10516,96 @@ }, "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 @@ -7972,8 +10622,8 @@ "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 @@ -7982,8 +10632,8 @@ "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" }, @@ -7991,8 +10641,8 @@ "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 @@ -8001,8 +10651,8 @@ "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 @@ -8011,8 +10661,8 @@ "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 @@ -8021,8 +10671,8 @@ "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 @@ -8031,8 +10681,8 @@ "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 @@ -8041,8 +10691,8 @@ "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 @@ -8051,8 +10701,8 @@ "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 @@ -8061,8 +10711,8 @@ "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" }, @@ -8070,8 +10720,8 @@ "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" }, @@ -8079,8 +10729,8 @@ "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" }, @@ -8088,36 +10738,36 @@ "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, + "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, + "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, + "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, + "input_cost_per_token": 5.625e-05, + "output_cost_per_token": 6.1875e-05, "litellm_provider": "aleph_alpha", "mode": "chat" }, @@ -8139,8 +10789,8 @@ "max_tokens": 8191, "max_input_tokens": 8191, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000125, - "output_cost_per_token": 0.0000125, + "input_cost_per_token": 1.25e-05, + "output_cost_per_token": 1.25e-05, "litellm_provider": "bedrock", "mode": "chat" }, @@ -8148,8 +10798,8 @@ "max_tokens": 8191, "max_input_tokens": 8191, "max_output_tokens": 8191, - "input_cost_per_token": 0.0000188, - "output_cost_per_token": 0.0000188, + "input_cost_per_token": 1.88e-05, + "output_cost_per_token": 1.88e-05, "litellm_provider": "bedrock", "mode": "chat" }, @@ -8157,8 +10807,8 @@ "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 @@ -8167,8 +10817,8 @@ "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" }, @@ -8176,8 +10826,8 @@ "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" }, @@ -8198,8 +10848,8 @@ "max_tokens": 4000, "max_input_tokens": 42000, "max_output_tokens": 4000, - "input_cost_per_token": 0.0000003, - "output_cost_per_token": 0.0000004, + "input_cost_per_token": 3e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "bedrock", "mode": "chat" }, @@ -8207,8 +10857,8 @@ "max_tokens": 8000, "max_input_tokens": 42000, "max_output_tokens": 8000, - "input_cost_per_token": 0.0000013, - "output_cost_per_token": 0.0000017, + "input_cost_per_token": 1.3e-06, + "output_cost_per_token": 1.7e-06, "litellm_provider": "bedrock", "mode": "chat" }, @@ -8216,8 +10866,8 @@ "max_tokens": 32000, "max_input_tokens": 42000, "max_output_tokens": 32000, - "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" }, @@ -8225,7 +10875,7 @@ "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", "mode": "embedding" @@ -8234,7 +10884,7 @@ "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", "mode": "embedding" @@ -8243,8 +10893,8 @@ "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", "supports_image_input": true, @@ -8259,8 +10909,8 @@ "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 @@ -8269,8 +10919,8 @@ "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 @@ -8279,8 +10929,8 @@ "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, @@ -8290,8 +10940,8 @@ "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, @@ -8301,8 +10951,8 @@ "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, @@ -8312,8 +10962,8 @@ "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 @@ -8322,8 +10972,8 @@ "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 @@ -8332,8 +10982,8 @@ "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 @@ -8342,8 +10992,8 @@ "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 @@ -8352,8 +11002,8 @@ "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 @@ -8362,8 +11012,8 @@ "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 @@ -8372,8 +11022,8 @@ "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, @@ -8383,8 +11033,8 @@ "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, @@ -8394,19 +11044,19 @@ "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": 10000, "max_input_tokens": 300000, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000000035, - "output_cost_per_token": 0.00000014, + "max_output_tokens": 10000, + "input_cost_per_token": 3.5e-08, + "output_cost_per_token": 1.4e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -8414,11 +11064,11 @@ "supports_response_schema": true }, "us.amazon.nova-micro-v1:0": { - "max_tokens": 4096, + "max_tokens": 10000, "max_input_tokens": 300000, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000000035, - "output_cost_per_token": 0.00000014, + "max_output_tokens": 10000, + "input_cost_per_token": 3.5e-08, + "output_cost_per_token": 1.4e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -8426,11 +11076,11 @@ "supports_response_schema": true }, "eu.amazon.nova-micro-v1:0": { - "max_tokens": 4096, + "max_tokens": 10000, "max_input_tokens": 300000, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000000046, - "output_cost_per_token": 0.000000184, + "max_output_tokens": 10000, + "input_cost_per_token": 4.6e-08, + "output_cost_per_token": 1.84e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -8438,11 +11088,11 @@ "supports_response_schema": true }, "amazon.nova-lite-v1:0": { - "max_tokens": 4096, + "max_tokens": 10000, "max_input_tokens": 128000, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000006, - "output_cost_per_token": 0.00000024, + "max_output_tokens": 10000, + "input_cost_per_token": 6e-08, + "output_cost_per_token": 2.4e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -8452,11 +11102,11 @@ "supports_response_schema": true }, "us.amazon.nova-lite-v1:0": { - "max_tokens": 4096, + "max_tokens": 10000, "max_input_tokens": 128000, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000006, - "output_cost_per_token": 0.00000024, + "max_output_tokens": 10000, + "input_cost_per_token": 6e-08, + "output_cost_per_token": 2.4e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -8466,11 +11116,11 @@ "supports_response_schema": true }, "eu.amazon.nova-lite-v1:0": { - "max_tokens": 4096, + "max_tokens": 10000, "max_input_tokens": 128000, - "max_output_tokens": 4096, - "input_cost_per_token": 0.000000078, - "output_cost_per_token": 0.000000312, + "max_output_tokens": 10000, + "input_cost_per_token": 7.8e-08, + "output_cost_per_token": 3.12e-07, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -8480,11 +11130,11 @@ "supports_response_schema": true }, "amazon.nova-pro-v1:0": { - "max_tokens": 4096, + "max_tokens": 10000, "max_input_tokens": 300000, - "max_output_tokens": 4096, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.0000032, + "max_output_tokens": 10000, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 3.2e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -8494,11 +11144,11 @@ "supports_response_schema": true }, "us.amazon.nova-pro-v1:0": { - "max_tokens": 4096, + "max_tokens": 10000, "max_input_tokens": 300000, - "max_output_tokens": 4096, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.0000032, + "max_output_tokens": 10000, + "input_cost_per_token": 8e-07, + "output_cost_per_token": 3.2e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -8508,17 +11158,17 @@ "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": 10000, "max_input_tokens": 300000, - "max_output_tokens": 4096, - "input_cost_per_token": 0.00000105, - "output_cost_per_token": 0.0000042, + "max_output_tokens": 10000, + "input_cost_per_token": 1.05e-06, + "output_cost_per_token": 4.2e-06, "litellm_provider": "bedrock_converse", "mode": "chat", "supports_function_calling": true, @@ -8528,12 +11178,66 @@ "supports_response_schema": true, "source": "https://aws.amazon.com/bedrock/pricing/" }, + "apac.amazon.nova-micro-v1:0": { + "max_tokens": 10000, + "max_input_tokens": 300000, + "max_output_tokens": 10000, + "input_cost_per_token": 3.7e-08, + "output_cost_per_token": 1.48e-07, + "litellm_provider": "bedrock_converse", + "mode": "chat", + "supports_function_calling": true, + "supports_prompt_caching": true, + "supports_response_schema": true + }, + "apac.amazon.nova-lite-v1:0": { + "max_tokens": 10000, + "max_input_tokens": 128000, + "max_output_tokens": 10000, + "input_cost_per_token": 6.3e-08, + "output_cost_per_token": 2.52e-07, + "litellm_provider": "bedrock_converse", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true + }, + "apac.amazon.nova-pro-v1:0": { + "max_tokens": 10000, + "max_input_tokens": 300000, + "max_output_tokens": 10000, + "input_cost_per_token": 8.4e-07, + "output_cost_per_token": 3.36e-06, + "litellm_provider": "bedrock_converse", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true + }, + "us.amazon.nova-premier-v1:0": { + "max_tokens": 10000, + "max_input_tokens": 1000000, + "max_output_tokens": 10000, + "input_cost_per_token": 2.5e-06, + "output_cost_per_token": 1.25e-05, + "litellm_provider": "bedrock_converse", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "supports_pdf_input": true, + "supports_prompt_caching": false, + "supports_response_schema": true + }, "anthropic.claude-3-sonnet-20240229-v1:0": { "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, @@ -8546,8 +11250,8 @@ "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, @@ -8562,8 +11266,8 @@ "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, @@ -8572,12 +11276,67 @@ "supports_pdf_input": true, "supports_tool_choice": true }, + "anthropic.claude-opus-4-20250514-v1:0": { + "max_tokens": 32000, + "max_input_tokens": 200000, + "max_output_tokens": 32000, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, + "search_context_cost_per_query": { + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 + }, + "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, + "supports_vision": true, + "tool_use_system_prompt_tokens": 159, + "supports_assistant_prefill": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_computer_use": true + }, + "anthropic.claude-sonnet-4-20250514-v1:0": { + "max_tokens": 64000, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "search_context_cost_per_query": { + "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-06, + "cache_read_input_token_cost": 3e-07, + "litellm_provider": "bedrock_converse", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "tool_use_system_prompt_tokens": 159, + "supports_assistant_prefill": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_computer_use": true + }, "anthropic.claude-3-7-sonnet-20250219-v1:0": { + "supports_computer_use": true, "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, + "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, @@ -8585,15 +11344,19 @@ "supports_assistant_prefill": true, "supports_prompt_caching": true, "supports_response_schema": true, + "supports_pdf_input": true, "supports_reasoning": true, "supports_tool_choice": true }, "anthropic.claude-3-5-sonnet-20241022-v2:0": { + "supports_computer_use": true, "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, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -8608,8 +11371,8 @@ "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, @@ -8622,8 +11385,10 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.000004, + "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, @@ -8637,8 +11402,8 @@ "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, @@ -8650,8 +11415,8 @@ "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, @@ -8664,8 +11429,8 @@ "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, @@ -8675,11 +11440,14 @@ "supports_tool_choice": true }, "us.anthropic.claude-3-5-sonnet-20241022-v2:0": { + "supports_computer_use": true, "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, + "cache_creation_input_token_cost": 3.75e-06, + "cache_read_input_token_cost": 3e-07, "litellm_provider": "bedrock", "mode": "chat", "supports_function_calling": true, @@ -8691,11 +11459,14 @@ "supports_tool_choice": true }, "us.anthropic.claude-3-7-sonnet-20250219-v1:0": { + "supports_computer_use": true, "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, + "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, @@ -8703,15 +11474,68 @@ "supports_assistant_prefill": true, "supports_prompt_caching": true, "supports_response_schema": true, + "supports_pdf_input": true, "supports_tool_choice": true, "supports_reasoning": true }, + "us.anthropic.claude-opus-4-20250514-v1:0": { + "max_tokens": 32000, + "max_input_tokens": 200000, + "max_output_tokens": 32000, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, + "search_context_cost_per_query": { + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 + }, + "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, + "supports_vision": true, + "tool_use_system_prompt_tokens": 159, + "supports_assistant_prefill": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_computer_use": true + }, + "us.anthropic.claude-sonnet-4-20250514-v1:0": { + "max_tokens": 64000, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "search_context_cost_per_query": { + "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-06, + "cache_read_input_token_cost": 3e-07, + "litellm_provider": "bedrock_converse", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "tool_use_system_prompt_tokens": 159, + "supports_assistant_prefill": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_computer_use": true + }, "us.anthropic.claude-3-haiku-20240307-v1:0": { "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, @@ -8724,8 +11548,10 @@ "max_tokens": 8192, "max_input_tokens": 200000, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.000004, + "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, @@ -8739,8 +11565,8 @@ "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, @@ -8752,8 +11578,8 @@ "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, @@ -8766,8 +11592,8 @@ "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, @@ -8777,11 +11603,12 @@ "supports_tool_choice": true }, "eu.anthropic.claude-3-5-sonnet-20241022-v2:0": { + "supports_computer_use": true, "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, @@ -8792,12 +11619,30 @@ "supports_response_schema": true, "supports_tool_choice": true }, + "eu.anthropic.claude-3-7-sonnet-20250219-v1:0": { + "supports_computer_use": true, + "max_tokens": 8192, + "max_input_tokens": 200000, + "max_output_tokens": 8192, + "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_response_schema": true, + "supports_pdf_input": true, + "supports_tool_choice": true, + "supports_reasoning": true + }, "eu.anthropic.claude-3-haiku-20240307-v1:0": { "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, @@ -8806,12 +11651,151 @@ "supports_pdf_input": true, "supports_tool_choice": true }, + "eu.anthropic.claude-opus-4-20250514-v1:0": { + "max_tokens": 32000, + "max_input_tokens": 200000, + "max_output_tokens": 32000, + "input_cost_per_token": 1.5e-05, + "output_cost_per_token": 7.5e-05, + "search_context_cost_per_query": { + "search_context_size_low": 0.01, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.01 + }, + "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, + "supports_vision": true, + "tool_use_system_prompt_tokens": 159, + "supports_assistant_prefill": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_computer_use": true + }, + "eu.anthropic.claude-sonnet-4-20250514-v1:0": { + "max_tokens": 64000, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "search_context_cost_per_query": { + "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-06, + "cache_read_input_token_cost": 3e-07, + "litellm_provider": "bedrock_converse", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "tool_use_system_prompt_tokens": 159, + "supports_assistant_prefill": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_computer_use": true + }, + "apac.anthropic.claude-3-haiku-20240307-v1:0": { + "max_tokens": 4096, + "max_input_tokens": 200000, + "max_output_tokens": 4096, + "input_cost_per_token": 2.5e-07, + "output_cost_per_token": 1.25e-06, + "litellm_provider": "bedrock", + "mode": "chat", + "supports_function_calling": true, + "supports_response_schema": true, + "supports_vision": true, + "supports_pdf_input": true, + "supports_tool_choice": true + }, + "apac.anthropic.claude-3-sonnet-20240229-v1:0": { + "max_tokens": 4096, + "max_input_tokens": 200000, + "max_output_tokens": 4096, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "litellm_provider": "bedrock", + "mode": "chat", + "supports_function_calling": true, + "supports_response_schema": true, + "supports_vision": true, + "supports_pdf_input": true, + "supports_tool_choice": true + }, + "apac.anthropic.claude-3-5-sonnet-20240620-v1:0": { + "max_tokens": 4096, + "max_input_tokens": 200000, + "max_output_tokens": 4096, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "litellm_provider": "bedrock", + "mode": "chat", + "supports_function_calling": true, + "supports_response_schema": true, + "supports_vision": true, + "supports_pdf_input": true, + "supports_tool_choice": true + }, + "apac.anthropic.claude-3-5-sonnet-20241022-v2:0": { + "max_tokens": 8192, + "max_input_tokens": 200000, + "max_output_tokens": 8192, + "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_assistant_prefill": true, + "supports_computer_use": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true + }, + "apac.anthropic.claude-sonnet-4-20250514-v1:0": { + "max_tokens": 64000, + "max_input_tokens": 200000, + "max_output_tokens": 64000, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 1.5e-05, + "search_context_cost_per_query": { + "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-06, + "cache_read_input_token_cost": 3e-07, + "litellm_provider": "bedrock_converse", + "mode": "chat", + "supports_function_calling": true, + "supports_vision": true, + "tool_use_system_prompt_tokens": 159, + "supports_assistant_prefill": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_reasoning": true, + "supports_computer_use": true + }, "eu.anthropic.claude-3-5-haiku-20241022-v1:0": { "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, @@ -8825,8 +11809,8 @@ "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, @@ -8838,8 +11822,8 @@ "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": "bedrock", "mode": "chat" }, @@ -8847,8 +11831,8 @@ "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": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -8857,8 +11841,8 @@ "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": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -8867,8 +11851,8 @@ "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": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -8895,8 +11879,8 @@ "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": "bedrock", "mode": "chat" }, @@ -8958,8 +11942,8 @@ "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": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -8968,8 +11952,8 @@ "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": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -8978,8 +11962,8 @@ "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": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -8988,8 +11972,8 @@ "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": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9018,8 +12002,8 @@ "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": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9088,8 +12072,8 @@ "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": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9098,8 +12082,8 @@ "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": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9108,8 +12092,8 @@ "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": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9118,8 +12102,8 @@ "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": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9148,8 +12132,8 @@ "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": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9218,8 +12202,8 @@ "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 @@ -9228,8 +12212,8 @@ "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 @@ -9278,8 +12262,8 @@ "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 @@ -9288,8 +12272,8 @@ "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 @@ -9318,8 +12302,8 @@ "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 @@ -9361,8 +12345,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.0000020, + "input_cost_per_token": 1.5e-06, + "output_cost_per_token": 2e-06, "litellm_provider": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9391,8 +12375,8 @@ "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": "bedrock", "mode": "chat", "supports_tool_choice": true @@ -9421,8 +12405,8 @@ "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 @@ -9431,8 +12415,8 @@ "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 @@ -9440,8 +12424,8 @@ "cohere.embed-english-v3": { "max_tokens": 512, "max_input_tokens": 512, - "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": "bedrock", "mode": "embedding", "supports_embedding_image_input": true @@ -9449,8 +12433,8 @@ "cohere.embed-multilingual-v3": { "max_tokens": 512, "max_input_tokens": 512, - "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": "bedrock", "mode": "embedding", "supports_embedding_image_input": true @@ -9459,21 +12443,20 @@ "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_tool_choice": false - }, "meta.llama3-3-70b-instruct-v1:0": { "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, @@ -9483,8 +12466,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000075, - "output_cost_per_token": 0.000001, + "input_cost_per_token": 7.5e-07, + "output_cost_per_token": 1e-06, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9492,8 +12475,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000195, - "output_cost_per_token": 0.00000256, + "input_cost_per_token": 1.95e-06, + "output_cost_per_token": 2.56e-06, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9501,8 +12484,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "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": "bedrock", "mode": "chat" }, @@ -9510,8 +12493,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "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": "bedrock", "mode": "chat" }, @@ -9519,8 +12502,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "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": "bedrock", "mode": "chat" }, @@ -9528,8 +12511,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000036, - "output_cost_per_token": 0.00000072, + "input_cost_per_token": 3.6e-07, + "output_cost_per_token": 7.2e-07, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9537,8 +12520,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000035, - "output_cost_per_token": 0.00000069, + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 6.9e-07, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9546,8 +12529,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000032, - "output_cost_per_token": 0.00000065, + "input_cost_per_token": 3.2e-07, + "output_cost_per_token": 6.5e-07, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9555,8 +12538,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000039, - "output_cost_per_token": 0.00000078, + "input_cost_per_token": 3.9e-07, + "output_cost_per_token": 7.8e-07, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9564,8 +12547,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.0000005, - "output_cost_per_token": 0.00000101, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 1.01e-06, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9573,8 +12556,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000265, - "output_cost_per_token": 0.0000035, + "input_cost_per_token": 2.65e-06, + "output_cost_per_token": 3.5e-06, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9582,8 +12565,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000265, - "output_cost_per_token": 0.0000035, + "input_cost_per_token": 2.65e-06, + "output_cost_per_token": 3.5e-06, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9591,8 +12574,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000265, - "output_cost_per_token": 0.0000035, + "input_cost_per_token": 2.65e-06, + "output_cost_per_token": 3.5e-06, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9600,8 +12583,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000318, - "output_cost_per_token": 0.0000042, + "input_cost_per_token": 3.18e-06, + "output_cost_per_token": 4.2e-06, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9609,8 +12592,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000305, - "output_cost_per_token": 0.00000403, + "input_cost_per_token": 3.05e-06, + "output_cost_per_token": 4.03e-06, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9618,8 +12601,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000286, - "output_cost_per_token": 0.00000378, + "input_cost_per_token": 2.86e-06, + "output_cost_per_token": 3.78e-06, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9627,8 +12610,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000345, - "output_cost_per_token": 0.00000455, + "input_cost_per_token": 3.45e-06, + "output_cost_per_token": 4.55e-06, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9636,8 +12619,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "max_output_tokens": 8192, - "input_cost_per_token": 0.00000445, - "output_cost_per_token": 0.00000588, + "input_cost_per_token": 4.45e-06, + "output_cost_per_token": 5.88e-06, "litellm_provider": "bedrock", "mode": "chat" }, @@ -9645,8 +12628,8 @@ "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, @@ -9656,8 +12639,8 @@ "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, @@ -9667,8 +12650,8 @@ "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, @@ -9678,8 +12661,8 @@ "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, @@ -9689,8 +12672,8 @@ "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, @@ -9700,8 +12683,8 @@ "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, @@ -9711,8 +12694,8 @@ "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, @@ -9722,8 +12705,8 @@ "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, @@ -9733,8 +12716,8 @@ "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, @@ -9744,8 +12727,8 @@ "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, @@ -9755,8 +12738,8 @@ "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, @@ -9766,8 +12749,8 @@ "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, @@ -9777,8 +12760,8 @@ "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, @@ -9789,8 +12772,8 @@ "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, @@ -9801,8 +12784,8 @@ "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, @@ -9813,8 +12796,8 @@ "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, @@ -9825,13 +12808,97 @@ "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_tool_choice": false }, + "meta.llama4-maverick-17b-instruct-v1:0": { + "max_tokens": 4096, + "max_input_tokens": 128000, + "max_output_tokens": 4096, + "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_tool_choice": false, + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] + }, + "us.meta.llama4-maverick-17b-instruct-v1:0": { + "max_tokens": 4096, + "max_input_tokens": 128000, + "max_output_tokens": 4096, + "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_tool_choice": false, + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] + }, + "meta.llama4-scout-17b-instruct-v1:0": { + "max_tokens": 4096, + "max_input_tokens": 128000, + "max_output_tokens": 4096, + "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_tool_choice": false, + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text", + "code" + ] + }, + "us.meta.llama4-scout-17b-instruct-v1:0": { + "max_tokens": 4096, + "max_input_tokens": 128000, + "max_output_tokens": 4096, + "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_tool_choice": false, + "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, @@ -9920,8 +12987,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "completion" }, @@ -9929,8 +12996,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "chat" }, @@ -9938,8 +13005,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "completion" }, @@ -9947,8 +13014,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "chat" }, @@ -9956,8 +13023,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, "litellm_provider": "sagemaker", "mode": "completion" }, @@ -9965,63 +13032,63 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.000, - "output_cost_per_token": 0.000, + "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, @@ -10030,8 +13097,8 @@ "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, @@ -10040,8 +13107,8 @@ "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, @@ -10049,8 +13116,8 @@ "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, @@ -10069,8 +13136,8 @@ "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, @@ -10093,6 +13160,55 @@ "mode": "chat", "supports_tool_choice": true }, + "together_ai/meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8": { + "litellm_provider": "together_ai", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "mode": "chat", + "supports_tool_choice": true + }, + "together_ai/meta-llama/Llama-4-Scout-17B-16E-Instruct": { + "litellm_provider": "together_ai", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "mode": "chat", + "supports_tool_choice": true + }, + "together_ai/meta-llama/Llama-3.2-3B-Instruct-Turbo": { + "litellm_provider": "together_ai", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "mode": "chat", + "supports_tool_choice": true + }, + "together_ai/Qwen/Qwen2.5-7B-Instruct-Turbo": { + "litellm_provider": "together_ai", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "mode": "chat", + "supports_tool_choice": true + }, + "together_ai/Qwen/Qwen2.5-72B-Instruct-Turbo": { + "litellm_provider": "together_ai", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "mode": "chat", + "supports_tool_choice": true + }, + "together_ai/deepseek-ai/DeepSeek-V3": { + "litellm_provider": "together_ai", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "mode": "chat", + "supports_tool_choice": true + }, + "together_ai/mistralai/Mistral-Small-24B-Instruct-2501": { + "litellm_provider": "together_ai", + "supports_function_calling": true, + "supports_parallel_function_calling": true, + "mode": "chat", + "supports_tool_choice": true + }, "ollama/codegemma": { "max_tokens": 8192, "max_input_tokens": 8192, @@ -10251,7 +13367,8 @@ "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": { "max_tokens": 8192, @@ -10260,7 +13377,8 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "completion" + "mode": "completion", + "supports_function_calling": true }, "ollama/mistral-7B-Instruct-v0.1": { "max_tokens": 8192, @@ -10269,7 +13387,8 @@ "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-7B-Instruct-v0.2": { "max_tokens": 32768, @@ -10278,7 +13397,8 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "chat" + "mode": "chat", + "supports_function_calling": true }, "ollama/mixtral-8x7B-Instruct-v0.1": { "max_tokens": 32768, @@ -10287,7 +13407,8 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "chat" + "mode": "chat", + "supports_function_calling": true }, "ollama/mixtral-8x22B-Instruct-v0.1": { "max_tokens": 65536, @@ -10296,7 +13417,8 @@ "input_cost_per_token": 0.0, "output_cost_per_token": 0.0, "litellm_provider": "ollama", - "mode": "chat" + "mode": "chat", + "supports_function_calling": true }, "ollama/codellama": { "max_tokens": 4096, @@ -10329,8 +13451,8 @@ "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 @@ -10339,8 +13461,8 @@ "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 @@ -10349,8 +13471,8 @@ "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 @@ -10359,8 +13481,8 @@ "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 @@ -10369,8 +13491,8 @@ "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 @@ -10379,8 +13501,8 @@ "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 @@ -10389,8 +13511,8 @@ "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" }, @@ -10398,8 +13520,8 @@ "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 @@ -10408,8 +13530,8 @@ "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 @@ -10418,8 +13540,8 @@ "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 @@ -10428,8 +13550,8 @@ "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 @@ -10438,8 +13560,8 @@ "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" }, @@ -10447,8 +13569,8 @@ "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 @@ -10457,8 +13579,8 @@ "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 @@ -10467,8 +13589,8 @@ "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 @@ -10477,8 +13599,8 @@ "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 @@ -10487,8 +13609,8 @@ "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 @@ -10497,8 +13619,8 @@ "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 @@ -10507,8 +13629,8 @@ "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, @@ -10519,8 +13641,8 @@ "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" }, @@ -10528,8 +13650,8 @@ "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 @@ -10538,8 +13660,8 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000035, - "output_cost_per_token": 0.00000140, + "input_cost_per_token": 3.5e-07, + "output_cost_per_token": 1.4e-06, "litellm_provider": "perplexity", "mode": "chat" }, @@ -10547,8 +13669,8 @@ "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000280, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 2.8e-06, "litellm_provider": "perplexity", "mode": "chat" }, @@ -10556,8 +13678,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "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": "perplexity", "mode": "chat" }, @@ -10565,8 +13687,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "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": "perplexity", "mode": "chat" }, @@ -10574,8 +13696,8 @@ "max_tokens": 127072, "max_input_tokens": 127072, "max_output_tokens": 127072, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000005, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 5e-06, "litellm_provider": "perplexity", "mode": "chat", "deprecation_date": "2025-02-22" @@ -10584,8 +13706,8 @@ "max_tokens": 127072, "max_input_tokens": 127072, "max_output_tokens": 127072, - "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": "perplexity", "mode": "chat", "deprecation_date": "2025-02-22" @@ -10594,8 +13716,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "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": "perplexity", "mode": "chat", "deprecation_date": "2025-02-22" @@ -10604,8 +13726,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "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": "perplexity", "mode": "chat", "deprecation_date": "2025-02-22" @@ -10614,54 +13736,18 @@ "max_tokens": 127072, "max_input_tokens": 127072, "max_output_tokens": 127072, - "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": "perplexity", - "mode": "chat" , + "mode": "chat", "deprecation_date": "2025-02-22" }, - "perplexity/sonar": { - "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", - "mode": "chat" - }, - "perplexity/sonar-pro": { - "max_tokens": 200000, - "max_input_tokens": 200000, - "max_output_tokens": 8096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "litellm_provider": "perplexity", - "mode": "chat" - }, - "perplexity/sonar": { - "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", - "mode": "chat" - }, - "perplexity/sonar-pro": { - "max_tokens": 200000, - "max_input_tokens": 200000, - "max_output_tokens": 8096, - "input_cost_per_token": 0.000003, - "output_cost_per_token": 0.000015, - "litellm_provider": "perplexity", - "mode": "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, + "input_cost_per_token": 7e-08, + "output_cost_per_token": 2.8e-07, "litellm_provider": "perplexity", "mode": "chat" }, @@ -10669,8 +13755,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000280, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 2.8e-06, "litellm_provider": "perplexity", "mode": "chat" }, @@ -10678,8 +13764,8 @@ "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" @@ -10688,8 +13774,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.0000000, - "output_cost_per_token": 0.00000280, + "input_cost_per_token": 0.0, + "output_cost_per_token": 2.8e-06, "input_cost_per_request": 0.005, "litellm_provider": "perplexity", "mode": "chat" @@ -10698,8 +13784,8 @@ "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "input_cost_per_token": 0.00000070, - "output_cost_per_token": 0.00000280, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 2.8e-06, "litellm_provider": "perplexity", "mode": "chat" }, @@ -10707,8 +13793,8 @@ "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" }, @@ -10716,8 +13802,8 @@ "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" }, @@ -10725,8 +13811,8 @@ "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" }, @@ -10735,7 +13821,7 @@ "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" @@ -10744,8 +13830,8 @@ "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" }, @@ -10754,73 +13840,149 @@ "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" }, + "perplexity/sonar": { + "max_tokens": 128000, + "max_input_tokens": 128000, + "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": 0.005, + "search_context_size_medium": 0.008, + "search_context_size_high": 0.012 + }, + "supports_web_search": true + }, + "perplexity/sonar-pro": { + "max_tokens": 8000, + "max_input_tokens": 200000, + "max_output_tokens": 8000, + "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": 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-06, + "output_cost_per_token": 5e-06, + "litellm_provider": "perplexity", + "mode": "chat", + "search_context_cost_per_query": { + "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 + }, + "perplexity/sonar-reasoning-pro": { + "max_tokens": 128000, + "max_input_tokens": 128000, + "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": 0.006, + "search_context_size_medium": 0.01, + "search_context_size_high": 0.014 + }, + "supports_web_search": true, + "supports_reasoning": true + }, + "perplexity/sonar-deep-research": { + "max_tokens": 128000, + "max_input_tokens": 128000, + "input_cost_per_token": 2e-06, + "output_cost_per_token": 8e-06, + "output_cost_per_reasoning_token": 3e-06, + "citation_cost_per_token": 2e-06, + "search_context_cost_per_query": { + "search_context_size_low": 0.005, + "search_context_size_medium": 0.005, + "search_context_size_high": 0.005 + }, + "litellm_provider": "perplexity", + "mode": "chat", + "supports_reasoning": true, + "supports_web_search": true + }, "fireworks_ai/accounts/fireworks/models/llama-v3p2-1b-instruct": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "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": "fireworks_ai", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": false, "supports_response_schema": true, "source": "https://fireworks.ai/pricing", - "supports_tool_choice": true + "supports_tool_choice": false }, "fireworks_ai/accounts/fireworks/models/llama-v3p2-3b-instruct": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "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": "fireworks_ai", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": false, "supports_response_schema": true, "source": "https://fireworks.ai/pricing", - "supports_tool_choice": true + "supports_tool_choice": false }, "fireworks_ai/accounts/fireworks/models/llama-v3p1-8b-instruct": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "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": "fireworks_ai", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": false, "supports_response_schema": true, "source": "https://fireworks.ai/pricing", - "supports_tool_choice": true + "supports_tool_choice": false }, "fireworks_ai/accounts/fireworks/models/llama-v3p2-11b-vision-instruct": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "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", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": false, "supports_vision": true, "supports_response_schema": true, "source": "https://fireworks.ai/pricing", - "supports_tool_choice": true + "supports_tool_choice": false }, - "accounts/fireworks/models/llama-v3p2-90b-vision-instruct": { + "fireworks_ai/accounts/fireworks/models/llama-v3p2-90b-vision-instruct": { "max_tokens": 16384, "max_input_tokens": 16384, "max_output_tokens": 16384, - "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_function_calling": true, + "supports_tool_choice": false, "supports_vision": true, "supports_response_schema": true, "source": "https://fireworks.ai/pricing" @@ -10829,8 +13991,8 @@ "max_tokens": 8192, "max_input_tokens": 8192, "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_function_calling": true, @@ -10842,8 +14004,8 @@ "max_tokens": 65536, "max_input_tokens": 65536, "max_output_tokens": 65536, - "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", "mode": "chat", "supports_function_calling": true, @@ -10855,72 +14017,144 @@ "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": "fireworks_ai", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": false, "supports_response_schema": true, "source": "https://fireworks.ai/pricing", - "supports_tool_choice": true + "supports_tool_choice": false }, "fireworks_ai/accounts/fireworks/models/qwen2p5-coder-32b-instruct": { "max_tokens": 4096, "max_input_tokens": 4096, "max_output_tokens": 4096, - "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_function_calling": true, + "supports_function_calling": false, "supports_response_schema": true, "source": "https://fireworks.ai/pricing", - "supports_tool_choice": true + "supports_tool_choice": false }, "fireworks_ai/accounts/fireworks/models/yi-large": { "max_tokens": 32768, "max_input_tokens": 32768, "max_output_tokens": 32768, - "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": "fireworks_ai", "mode": "chat", - "supports_function_calling": true, + "supports_function_calling": false, "supports_response_schema": true, "source": "https://fireworks.ai/pricing", - "supports_tool_choice": true + "supports_tool_choice": false }, "fireworks_ai/accounts/fireworks/models/deepseek-coder-v2-instruct": { "max_tokens": 65536, "max_input_tokens": 65536, - "max_output_tokens": 8192, - "input_cost_per_token": 0.0000012, - "output_cost_per_token": 0.0000012, + "max_output_tokens": 65536, + "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_function_calling": false, "supports_response_schema": true, "source": "https://fireworks.ai/pricing", - "supports_tool_choice": true + "supports_tool_choice": false }, "fireworks_ai/accounts/fireworks/models/deepseek-v3": { "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, "source": "https://fireworks.ai/pricing", - "supports_tool_choice": true + "supports_tool_choice": false + }, + "fireworks_ai/accounts/fireworks/models/deepseek-r1": { + "max_tokens": 20480, + "max_input_tokens": 128000, + "max_output_tokens": 20480, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 8e-06, + "litellm_provider": "fireworks_ai", + "mode": "chat", + "supports_response_schema": true, + "source": "https://fireworks.ai/pricing", + "supports_tool_choice": false + }, + "fireworks_ai/accounts/fireworks/models/deepseek-r1-basic": { + "max_tokens": 20480, + "max_input_tokens": 128000, + "max_output_tokens": 20480, + "input_cost_per_token": 5.5e-07, + "output_cost_per_token": 2.19e-06, + "litellm_provider": "fireworks_ai", + "mode": "chat", + "supports_response_schema": true, + "source": "https://fireworks.ai/pricing", + "supports_tool_choice": false + }, + "fireworks_ai/accounts/fireworks/models/deepseek-r1-0528": { + "max_tokens": 160000, + "max_input_tokens": 160000, + "max_output_tokens": 160000, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 8e-06, + "litellm_provider": "fireworks_ai", + "mode": "chat", + "source": "https://fireworks.ai/pricing", + "supports_tool_choice": false, + "supports_response_schema": true + }, + "fireworks_ai/accounts/fireworks/models/llama-v3p1-405b-instruct": { + "max_tokens": 16384, + "max_input_tokens": 128000, + "max_output_tokens": 16384, + "input_cost_per_token": 3e-06, + "output_cost_per_token": 3e-06, + "litellm_provider": "fireworks_ai", + "mode": "chat", + "supports_response_schema": true, + "source": "https://fireworks.ai/pricing", + "supports_tool_choice": true, + "supports_function_calling": true + }, + "fireworks_ai/accounts/fireworks/models/llama4-maverick-instruct-basic": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 2.2e-07, + "output_cost_per_token": 8.8e-07, + "litellm_provider": "fireworks_ai", + "mode": "chat", + "supports_response_schema": true, + "source": "https://fireworks.ai/pricing", + "supports_tool_choice": false + }, + "fireworks_ai/accounts/fireworks/models/llama4-scout-instruct-basic": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 1.5e-07, + "output_cost_per_token": 6e-07, + "litellm_provider": "fireworks_ai", + "mode": "chat", + "supports_response_schema": true, + "source": "https://fireworks.ai/pricing", + "supports_tool_choice": false }, - "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" @@ -10928,8 +14162,8 @@ "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" @@ -10937,8 +14171,8 @@ "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" @@ -10946,8 +14180,8 @@ "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" @@ -10955,30 +14189,35 @@ "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-16b": { - "input_cost_per_token": 0.0000002, - "output_cost_per_token": 0.0000002, + "fireworks-ai-up-to-4b": { + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "fireworks_ai" }, - "fireworks-ai-16.1b-to-80b": { - "input_cost_per_token": 0.0000009, - "output_cost_per_token": 0.0000009, + "fireworks-ai-4.1b-to-16b": { + "input_cost_per_token": 2e-07, + "output_cost_per_token": 2e-07, + "litellm_provider": "fireworks_ai" + }, + "fireworks-ai-above-16b": { + "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": { @@ -10987,270 +14226,270 @@ "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, + "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, + "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, + "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, + "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, + "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, + "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, + "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, + "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, + "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, + "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, + "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, + "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": { + "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, + "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": { + }, + "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, + "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": { + }, + "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, + "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": { + }, + "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, + "input_cost_per_token": 1.923e-06, + "output_cost_per_token": 1.923e-06, "litellm_provider": "cloudflare", "mode": "chat" - }, - "voyage/voyage-01": { + }, + "voyage/voyage-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-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" }, @@ -11259,8 +14498,8 @@ "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" @@ -11270,8 +14509,8 @@ "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" @@ -11280,14 +14519,16 @@ "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": 0.00017857, + "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, @@ -11297,241 +14538,371 @@ "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, + "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, + "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, + "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": 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)." + }, "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, + "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, + "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, + "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, + "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, + "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, + "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, + "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, + "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": 16000, - "max_input_tokens": 16000, - "max_output_tokens": 16000, - "input_cost_per_token": 0.0000001, - "output_cost_per_token": 0.0000002, + "max_tokens": 16384, + "max_input_tokens": 16384, + "max_output_tokens": 16384, + "input_cost_per_token": 1e-07, + "output_cost_per_token": 2e-07, "litellm_provider": "sambanova", - "supports_function_calling": true, "mode": "chat", - "supports_tool_choice": true - }, - "sambanova/Meta-Llama-3.1-70B-Instruct": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 0.0000006, - "output_cost_per_token": 0.0000012, - "litellm_provider": "sambanova", "supports_function_calling": true, - "mode": "chat", - "supports_tool_choice": true + "supports_tool_choice": true, + "supports_response_schema": true, + "source": "https://cloud.sambanova.ai/plans/pricing" }, "sambanova/Meta-Llama-3.1-405B-Instruct": { - "max_tokens": 16000, - "max_input_tokens": 16000, - "max_output_tokens": 16000, - "input_cost_per_token": 0.000005, - "output_cost_per_token": 0.000010, + "max_tokens": 16384, + "max_input_tokens": 16384, + "max_output_tokens": 16384, + "input_cost_per_token": 5e-06, + "output_cost_per_token": 1e-05, "litellm_provider": "sambanova", - "supports_function_calling": true, "mode": "chat", - "supports_tool_choice": true + "supports_function_calling": true, + "supports_tool_choice": true, + "supports_response_schema": true, + "source": "https://cloud.sambanova.ai/plans/pricing" }, "sambanova/Meta-Llama-3.2-1B-Instruct": { - "max_tokens": 16000, - "max_input_tokens": 16000, - "max_output_tokens": 16000, - "input_cost_per_token": 0.0000004, - "output_cost_per_token": 0.0000008, + "max_tokens": 16384, + "max_input_tokens": 16384, + "max_output_tokens": 16384, + "input_cost_per_token": 4e-08, + "output_cost_per_token": 8e-08, "litellm_provider": "sambanova", - "supports_function_calling": true, "mode": "chat", - "supports_tool_choice": true + "source": "https://cloud.sambanova.ai/plans/pricing" }, "sambanova/Meta-Llama-3.2-3B-Instruct": { - "max_tokens": 4000, - "max_input_tokens": 4000, - "max_output_tokens": 4000, - "input_cost_per_token": 0.0000008, - "output_cost_per_token": 0.0000016, + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 8e-08, + "output_cost_per_token": 1.6e-07, "litellm_provider": "sambanova", - "supports_function_calling": true, "mode": "chat", - "supports_tool_choice": true + "source": "https://cloud.sambanova.ai/plans/pricing" }, - "sambanova/Qwen2.5-Coder-32B-Instruct": { - "max_tokens": 8000, - "max_input_tokens": 8000, - "max_output_tokens": 8000, - "input_cost_per_token": 0.0000015, - "output_cost_per_token": 0.000003, + "sambanova/Llama-4-Maverick-17B-128E-Instruct": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "max_output_tokens": 131072, + "input_cost_per_token": 6.3e-07, + "output_cost_per_token": 1.8e-06, "litellm_provider": "sambanova", - "supports_function_calling": true, "mode": "chat", - "supports_tool_choice": true + "supports_function_calling": true, + "supports_tool_choice": true, + "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" + } }, - "sambanova/Qwen2.5-72B-Instruct": { - "max_tokens": 8000, - "max_input_tokens": 8000, - "max_output_tokens": 8000, - "input_cost_per_token": 0.000002, - "output_cost_per_token": 0.000004, + "sambanova/Llama-4-Scout-17B-16E-Instruct": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "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" + } + }, + "sambanova/Meta-Llama-3.3-70B-Instruct": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "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, + "supports_response_schema": true, + "supports_tool_choice": true, + "source": "https://cloud.sambanova.ai/plans/pricing" + }, + "sambanova/Meta-Llama-Guard-3-8B": { + "max_tokens": 16384, + "max_input_tokens": 16384, + "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" + }, + "sambanova/Qwen3-32B": { + "max_tokens": 8192, + "max_input_tokens": 8192, + "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, + "supports_reasoning": true, "mode": "chat", - "supports_tool_choice": true + "source": "https://cloud.sambanova.ai/plans/pricing" + }, + "sambanova/QwQ-32B": { + "max_tokens": 16384, + "max_input_tokens": 16384, + "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" + }, + "sambanova/Qwen2-Audio-7B-Instruct": { + "max_tokens": 4096, + "max_input_tokens": 4096, + "max_output_tokens": 4096, + "input_cost_per_token": 5e-07, + "output_cost_per_token": 0.0001, + "litellm_provider": "sambanova", + "mode": "chat", + "supports_audio_input": true, + "source": "https://cloud.sambanova.ai/plans/pricing" + }, + "sambanova/DeepSeek-R1-Distill-Llama-70B": { + "max_tokens": 131072, + "max_input_tokens": 131072, + "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" + }, + "sambanova/DeepSeek-R1": { + "max_tokens": 32768, + "max_input_tokens": 32768, + "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" + }, + "sambanova/DeepSeek-V3-0324": { + "max_tokens": 32768, + "max_input_tokens": 32768, + "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, + "supports_tool_choice": true, + "supports_reasoning": true, + "source": "https://cloud.sambanova.ai/plans/pricing" }, "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": { @@ -11539,8 +14910,8 @@ "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" }, @@ -11560,6 +14931,7 @@ "mode": "chat" }, "snowflake/claude-3-5-sonnet": { + "supports_computer_use": true, "max_tokens": 18000, "max_input_tokens": 18000, "max_output_tokens": 8192, @@ -11752,5 +15124,692 @@ "mode": "chat", "supported_endpoints": ["/v1/chat/completions"], "supported_modalities": ["text"] + }, + "nscale/meta-llama/Llama-4-Scout-17B-16E-Instruct": { + "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-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-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-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-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-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", + "metadata": { + "notes": "Pricing listed as $0.75/1M tokens total. Assumed 50/50 split for input/output." + } + }, + "nscale/deepseek-ai/DeepSeek-R1-Distill-Llama-8B": { + "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", + "metadata": { + "notes": "Pricing listed as $0.05/1M tokens total. Assumed 50/50 split for input/output." + } + }, + "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": { + "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", + "metadata": { + "notes": "Pricing listed as $0.18/1M tokens total. Assumed 50/50 split for input/output." + } + }, + "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": { + "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", + "metadata": { + "notes": "Pricing listed as $0.40/1M tokens total. Assumed 50/50 split for input/output." + } + }, + "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": { + "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", + "metadata": { + "notes": "Pricing listed as $0.14/1M tokens total. Assumed 50/50 split for input/output." + } + }, + "nscale/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": { + "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", + "metadata": { + "notes": "Pricing listed as $0.30/1M tokens total. Assumed 50/50 split for input/output." + } + }, + "nscale/mistralai/mixtral-8x22b-instruct-v0.1": { + "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", + "metadata": { + "notes": "Pricing listed as $1.20/1M tokens total. Assumed 50/50 split for input/output." + } + }, + "nscale/meta-llama/Llama-3.1-8B-Instruct": { + "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", + "metadata": { + "notes": "Pricing listed as $0.06/1M tokens total. Assumed 50/50 split for input/output." + } + }, + "nscale/meta-llama/Llama-3.3-70B-Instruct": { + "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", + "metadata": { + "notes": "Pricing listed as $0.40/1M tokens total. Assumed 50/50 split for input/output." + } + }, + "nscale/black-forest-labs/FLUX.1-schnell": { + "mode": "image_generation", + "input_cost_per_pixel": 1.3e-09, + "output_cost_per_pixel": 0.0, + "litellm_provider": "nscale", + "supported_endpoints": [ + "/v1/images/generations" + ], + "source": "https://docs.nscale.com/docs/inference/serverless-models/current#image-models" + }, + "nscale/stabilityai/stable-diffusion-xl-base-1.0": { + "mode": "image_generation", + "input_cost_per_pixel": 3e-09, + "output_cost_per_pixel": 0.0, + "litellm_provider": "nscale", + "supported_endpoints": [ + "/v1/images/generations" + ], + "source": "https://docs.nscale.com/docs/inference/serverless-models/current#image-models" + }, + "featherless_ai/featherless-ai/Qwerky-72B": { + "max_tokens": 32768, + "max_input_tokens": 32768, + "max_output_tokens": 4096, + "litellm_provider": "featherless_ai", + "mode": "chat" + }, + "featherless_ai/featherless-ai/Qwerky-QwQ-32B": { + "max_tokens": 32768, + "max_input_tokens": 32768, + "max_output_tokens": 4096, + "litellm_provider": "featherless_ai", + "mode": "chat" + }, + "deepgram/nova-3": { + "mode": "audio_transcription", + "input_cost_per_second": 7.167e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0043, + "calculation": "$0.0043/60 seconds = $0.00007167 per second" + } + }, + "deepgram/nova-3-general": { + "mode": "audio_transcription", + "input_cost_per_second": 7.167e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0043, + "calculation": "$0.0043/60 seconds = $0.00007167 per second" + } + }, + "deepgram/nova-3-medical": { + "mode": "audio_transcription", + "input_cost_per_second": 8.667e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0052, + "calculation": "$0.0052/60 seconds = $0.00008667 per second (multilingual)" + } + }, + "deepgram/nova-2": { + "mode": "audio_transcription", + "input_cost_per_second": 7.167e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0043, + "calculation": "$0.0043/60 seconds = $0.00007167 per second" + } + }, + "deepgram/nova-2-general": { + "mode": "audio_transcription", + "input_cost_per_second": 7.167e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0043, + "calculation": "$0.0043/60 seconds = $0.00007167 per second" + } + }, + "deepgram/nova-2-meeting": { + "mode": "audio_transcription", + "input_cost_per_second": 7.167e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0043, + "calculation": "$0.0043/60 seconds = $0.00007167 per second" + } + }, + "deepgram/nova-2-phonecall": { + "mode": "audio_transcription", + "input_cost_per_second": 7.167e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0043, + "calculation": "$0.0043/60 seconds = $0.00007167 per second" + } + }, + "deepgram/nova-2-voicemail": { + "mode": "audio_transcription", + "input_cost_per_second": 7.167e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0043, + "calculation": "$0.0043/60 seconds = $0.00007167 per second" + } + }, + "deepgram/nova-2-finance": { + "mode": "audio_transcription", + "input_cost_per_second": 7.167e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0043, + "calculation": "$0.0043/60 seconds = $0.00007167 per second" + } + }, + "deepgram/nova-2-conversationalai": { + "mode": "audio_transcription", + "input_cost_per_second": 7.167e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0043, + "calculation": "$0.0043/60 seconds = $0.00007167 per second" + } + }, + "deepgram/nova-2-video": { + "mode": "audio_transcription", + "input_cost_per_second": 7.167e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0043, + "calculation": "$0.0043/60 seconds = $0.00007167 per second" + } + }, + "deepgram/nova-2-drivethru": { + "mode": "audio_transcription", + "input_cost_per_second": 7.167e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0043, + "calculation": "$0.0043/60 seconds = $0.00007167 per second" + } + }, + "deepgram/nova-2-automotive": { + "mode": "audio_transcription", + "input_cost_per_second": 7.167e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0043, + "calculation": "$0.0043/60 seconds = $0.00007167 per second" + } + }, + "deepgram/nova-2-atc": { + "mode": "audio_transcription", + "input_cost_per_second": 7.167e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0043, + "calculation": "$0.0043/60 seconds = $0.00007167 per second" + } + }, + "deepgram/nova": { + "mode": "audio_transcription", + "input_cost_per_second": 7.167e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0043, + "calculation": "$0.0043/60 seconds = $0.00007167 per second" + } + }, + "deepgram/nova-general": { + "mode": "audio_transcription", + "input_cost_per_second": 7.167e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0043, + "calculation": "$0.0043/60 seconds = $0.00007167 per second" + } + }, + "deepgram/nova-phonecall": { + "mode": "audio_transcription", + "input_cost_per_second": 7.167e-05, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0043, + "calculation": "$0.0043/60 seconds = $0.00007167 per second" + } + }, + "deepgram/enhanced": { + "mode": "audio_transcription", + "input_cost_per_second": 0.00024167, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0145, + "calculation": "$0.0145/60 seconds = $0.00024167 per second" + } + }, + "deepgram/enhanced-general": { + "mode": "audio_transcription", + "input_cost_per_second": 0.00024167, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0145, + "calculation": "$0.0145/60 seconds = $0.00024167 per second" + } + }, + "deepgram/enhanced-meeting": { + "mode": "audio_transcription", + "input_cost_per_second": 0.00024167, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0145, + "calculation": "$0.0145/60 seconds = $0.00024167 per second" + } + }, + "deepgram/enhanced-phonecall": { + "mode": "audio_transcription", + "input_cost_per_second": 0.00024167, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0145, + "calculation": "$0.0145/60 seconds = $0.00024167 per second" + } + }, + "deepgram/enhanced-finance": { + "mode": "audio_transcription", + "input_cost_per_second": 0.00024167, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0145, + "calculation": "$0.0145/60 seconds = $0.00024167 per second" + } + }, + "deepgram/base": { + "mode": "audio_transcription", + "input_cost_per_second": 0.00020833, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0125, + "calculation": "$0.0125/60 seconds = $0.00020833 per second" + } + }, + "deepgram/base-general": { + "mode": "audio_transcription", + "input_cost_per_second": 0.00020833, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0125, + "calculation": "$0.0125/60 seconds = $0.00020833 per second" + } + }, + "deepgram/base-meeting": { + "mode": "audio_transcription", + "input_cost_per_second": 0.00020833, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0125, + "calculation": "$0.0125/60 seconds = $0.00020833 per second" + } + }, + "deepgram/base-phonecall": { + "mode": "audio_transcription", + "input_cost_per_second": 0.00020833, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0125, + "calculation": "$0.0125/60 seconds = $0.00020833 per second" + } + }, + "deepgram/base-voicemail": { + "mode": "audio_transcription", + "input_cost_per_second": 0.00020833, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0125, + "calculation": "$0.0125/60 seconds = $0.00020833 per second" + } + }, + "deepgram/base-finance": { + "mode": "audio_transcription", + "input_cost_per_second": 0.00020833, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0125, + "calculation": "$0.0125/60 seconds = $0.00020833 per second" + } + }, + "deepgram/base-conversationalai": { + "mode": "audio_transcription", + "input_cost_per_second": 0.00020833, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0125, + "calculation": "$0.0125/60 seconds = $0.00020833 per second" + } + }, + "deepgram/base-video": { + "mode": "audio_transcription", + "input_cost_per_second": 0.00020833, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "original_pricing_per_minute": 0.0125, + "calculation": "$0.0125/60 seconds = $0.00020833 per second" + } + }, + "deepgram/whisper": { + "mode": "audio_transcription", + "input_cost_per_second": 0.0001, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "notes": "Deepgram's hosted OpenAI Whisper models - pricing may differ from native Deepgram models" + } + }, + "deepgram/whisper-tiny": { + "mode": "audio_transcription", + "input_cost_per_second": 0.0001, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "notes": "Deepgram's hosted OpenAI Whisper models - pricing may differ from native Deepgram models" + } + }, + "deepgram/whisper-base": { + "mode": "audio_transcription", + "input_cost_per_second": 0.0001, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "notes": "Deepgram's hosted OpenAI Whisper models - pricing may differ from native Deepgram models" + } + }, + "deepgram/whisper-small": { + "mode": "audio_transcription", + "input_cost_per_second": 0.0001, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "notes": "Deepgram's hosted OpenAI Whisper models - pricing may differ from native Deepgram models" + } + }, + "deepgram/whisper-medium": { + "mode": "audio_transcription", + "input_cost_per_second": 0.0001, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "notes": "Deepgram's hosted OpenAI Whisper models - pricing may differ from native Deepgram models" + } + }, + "deepgram/whisper-large": { + "mode": "audio_transcription", + "input_cost_per_second": 0.0001, + "output_cost_per_second": 0.0, + "litellm_provider": "deepgram", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://deepgram.com/pricing", + "metadata": { + "notes": "Deepgram's hosted OpenAI Whisper models - pricing may differ from native Deepgram models" + } + }, + "elevenlabs/scribe_v1": { + "mode": "audio_transcription", + "input_cost_per_second": 0.0000611, + "output_cost_per_second": 0.0, + "litellm_provider": "elevenlabs", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://elevenlabs.io/pricing", + "metadata": { + "original_pricing_per_hour": 0.22, + "calculation": "$0.22/hour = $0.00366/minute = $0.0000611 per second (enterprise pricing)", + "notes": "ElevenLabs Scribe v1 - state-of-the-art speech recognition model with 99 language support" + } + }, + "elevenlabs/scribe_v1_experimental": { + "mode": "audio_transcription", + "input_cost_per_second": 0.0000611, + "output_cost_per_second": 0.0, + "litellm_provider": "elevenlabs", + "supported_endpoints": [ + "/v1/audio/transcriptions" + ], + "source": "https://elevenlabs.io/pricing", + "metadata": { + "original_pricing_per_hour": 0.22, + "calculation": "$0.22/hour = $0.00366/minute = $0.0000611 per second (enterprise pricing)", + "notes": "ElevenLabs Scribe v1 experimental - enhanced version of the main Scribe model" + } } -} +} \ No newline at end of file