diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index d8039c0d96e..157c79ca388 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -3,6 +3,8 @@ "cache_creation_input_audio_token_cost": 4e-07, "cache_read_input_audio_token_cost": 4e-07, "cache_read_input_token_cost": 4e-07, + "deprecation_date": "2027-06-25", + "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "input_cost_per_audio_token": 3.2e-05, "input_cost_per_image_token": 5e-06, "input_cost_per_token": 4e-06, @@ -42,6 +44,8 @@ "cache_creation_input_audio_token_cost": 3e-07, "cache_read_input_audio_token_cost": 3e-07, "cache_read_input_token_cost": 6e-08, + "deprecation_date": "2027-06-25", + "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "input_cost_per_audio_token": 1e-05, "input_cost_per_image_token": 8e-07, "input_cost_per_token": 6e-07, @@ -5348,7 +5352,7 @@ "azure/eu/gpt-5-mini-2025-08-07": { "cache_read_input_token_cost": 2.75e-08, "cache_read_input_token_cost_priority": 4.95e-08, - "deprecation_date": "2027-02-09", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/eu/gpt-5.4-mini", "input_cost_per_token": 2.75e-07, @@ -5392,7 +5396,7 @@ "cache_read_input_token_cost_batches": 1.375e-08 }, "azure/eu/gpt-5.1": { - "deprecation_date": "2027-05-15", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/eu/gpt-5.2", "cache_read_input_token_cost": 1.375e-07, @@ -5742,7 +5746,7 @@ "supports_vision": true }, "azure/global/gpt-5.1": { - "deprecation_date": "2027-05-15", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "cache_read_input_token_cost": 1.25e-07, "cache_read_input_token_cost_batches": 6.25e-08, @@ -6226,7 +6230,7 @@ "supports_vision": true }, "azure/gpt-4o-2024-05-13": { - "deprecation_date": "2026-12-09", + "deprecation_date": "2027-01-14", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/gpt-5.1", "input_cost_per_token": 5e-06, @@ -7044,7 +7048,7 @@ "azure/gpt-5.1-2025-11-13": { "cache_read_input_token_cost": 1.25e-07, "cache_read_input_token_cost_priority": 2.5e-07, - "deprecation_date": "2027-05-15", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/gpt-5.2", "input_cost_per_token": 1.25e-06, @@ -7287,7 +7291,7 @@ "supports_vision": true }, "azure/gpt-5-mini": { - "deprecation_date": "2027-02-09", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/gpt-5.4-mini", "cache_read_input_token_cost": 2.5e-08, @@ -7335,7 +7339,7 @@ "azure/gpt-5-mini-2025-08-07": { "cache_read_input_token_cost": 2.5e-08, "cache_read_input_token_cost_priority": 4.5e-08, - "deprecation_date": "2027-02-09", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/gpt-5.4-mini", "input_cost_per_token": 2.5e-07, @@ -7503,7 +7507,7 @@ "supports_vision": true }, "azure/gpt-5.1": { - "deprecation_date": "2027-05-15", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/gpt-5.2", "cache_read_input_token_cost": 1.25e-07, @@ -12384,7 +12388,7 @@ "azure/us/gpt-5-mini-2025-08-07": { "cache_read_input_token_cost": 2.75e-08, "cache_read_input_token_cost_priority": 4.95e-08, - "deprecation_date": "2027-02-09", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/us/gpt-5.4-mini", "input_cost_per_token": 2.75e-07, @@ -12471,7 +12475,7 @@ "cache_read_input_token_cost_batches": 2.75e-09 }, "azure/us/gpt-5.1": { - "deprecation_date": "2027-05-15", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/us/gpt-5.2", "cache_read_input_token_cost": 1.375e-07, @@ -31527,6 +31531,8 @@ "input_cost_per_audio_token_batches": 2.5e-07 }, "gemini/gemini-embedding-2-preview": { + "deprecation_date": "2026-08-10", + "deprecation_source": "https://ai.google.dev/gemini-api/docs/deprecations", "input_cost_per_audio_token": 6.5e-06, "input_cost_per_audio_token_batches": 3.25e-06, "input_cost_per_image_token": 4.5e-07, @@ -31542,6 +31548,7 @@ "output_vector_size": 3072, "rpm": 10000, "source": "https://ai.google.dev/gemini-api/docs/pricing", + "successor_model": "gemini/gemini-embedding-2", "supports_audio_input": true, "supports_embedding_image_input": true, "supports_multimodal": true, @@ -67587,11 +67594,12 @@ }, "xai/grok-imagine-image-quality": { "deprecation_date": "2026-11-02", - "deprecation_source": "https://docs.x.ai/docs/models", + "deprecation_source": "https://docs.x.ai/developers/migration/imagine-image-quality-nov-2", "input_cost_per_image": 0.05, "litellm_provider": "xai", "mode": "image_generation", "source": "https://docs.x.ai/docs/models", + "successor_model": "xai/grok-imagine-image-2.0", "supported_endpoints": [ "/v1/images/generations" ], @@ -67605,11 +67613,12 @@ }, "xai/grok-imagine-image-quality-20260403": { "deprecation_date": "2026-11-02", - "deprecation_source": "https://docs.x.ai/docs/models", + "deprecation_source": "https://docs.x.ai/developers/migration/imagine-image-quality-nov-2", "input_cost_per_image": 0.05, "litellm_provider": "xai", "mode": "image_generation", "source": "https://docs.x.ai/docs/models", + "successor_model": "xai/grok-imagine-image-2.0", "supported_endpoints": [ "/v1/images/generations" ], @@ -67623,11 +67632,12 @@ }, "xai/grok-imagine-image-quality-latest": { "deprecation_date": "2026-11-02", - "deprecation_source": "https://docs.x.ai/docs/models", + "deprecation_source": "https://docs.x.ai/developers/migration/imagine-image-quality-nov-2", "input_cost_per_image": 0.05, "litellm_provider": "xai", "mode": "image_generation", "source": "https://docs.x.ai/docs/models", + "successor_model": "xai/grok-imagine-image-2.0", "supported_endpoints": [ "/v1/images/generations" ], @@ -67641,11 +67651,12 @@ }, "xai/grok-imagine-image-pro": { "deprecation_date": "2026-05-15", - "deprecation_source": "https://docs.x.ai/docs/models", + "deprecation_source": "https://docs.x.ai/developers/migration/may-15-retirement", "input_cost_per_image": 0.05, "litellm_provider": "xai", "mode": "image_generation", "source": "https://docs.x.ai/docs/models", + "successor_model": "xai/grok-imagine-image-2.0", "supported_endpoints": [ "/v1/images/generations" ], @@ -67763,6 +67774,7 @@ }, "xai/grok-voice-transcribe-1.0": { "deprecation_date": "2026-10-02", + "deprecation_source": "https://docs.x.ai/docs/release-notes", "successor_model": "xai/grok-voice-transcribe-2.0", "input_cost_per_second": 2.778e-05, "litellm_provider": "xai", @@ -71226,7 +71238,7 @@ "source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'" }, "azure/eu/gpt-4o-2024-05-13": { - "deprecation_date": "2026-12-09", + "deprecation_date": "2027-01-14", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/eu/gpt-5.1", "input_cost_per_token": 5.5e-06, @@ -71273,7 +71285,7 @@ "source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'" }, "azure/eu/gpt-5-mini": { - "deprecation_date": "2027-02-09", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/eu/gpt-5.4-mini", "cache_read_input_token_cost": 2.75e-08, @@ -71846,7 +71858,7 @@ "source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'" }, "azure/us/gpt-4o-2024-05-13": { - "deprecation_date": "2026-12-09", + "deprecation_date": "2027-01-14", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/us/gpt-5.1", "input_cost_per_token": 5.5e-06, @@ -71893,7 +71905,7 @@ "source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'" }, "azure/us/gpt-5-mini": { - "deprecation_date": "2027-02-09", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/us/gpt-5.4-mini", "cache_read_input_token_cost": 2.75e-08, @@ -74556,6 +74568,7 @@ "cache_read_input_token_cost": 2e-07, "cache_read_input_token_cost_above_200k_tokens": 4e-07, "deprecation_date": "2026-05-15", + "deprecation_source": "https://docs.x.ai/developers/migration/may-15-retirement", "input_cost_per_image_token": 1e-06, "input_cost_per_token": 1e-06, "input_cost_per_token_above_200k_tokens": 2e-06, @@ -74566,6 +74579,7 @@ "output_cost_per_token": 2e-06, "output_cost_per_token_above_200k_tokens": 4e-06, "source": "https://api.x.ai/v1/language-models", + "successor_model": "xai/grok-build-0.1", "supports_function_calling": true, "supports_prompt_caching": true, "supports_reasoning": true, @@ -77927,6 +77941,7 @@ "litellm_provider": "xai", "mode": "realtime", "source": "https://docs.x.ai/developers/pricing", + "successor_model": "xai/grok-voice-3.0", "supported_endpoints": [ "/v1/realtime" ], @@ -82153,6 +82168,23 @@ "supports_tool_choice": true, "supports_vision": true }, + "fireworks_ai/accounts/fireworks/routers/deepseek-v4p1-flash-1017-ultrafast": { + "cache_read_input_token_cost": 4.8e-08, + "input_cost_per_token": 2.4e-06, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 1048576, + "max_output_tokens": 393216, + "mode": "chat", + "output_cost_per_token": 9.6e-06, + "source": "https://docs.fireworks.ai/serverless/pricing.md", + "supports_function_calling": true, + "supports_mid_conversation_system": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, "fireworks_ai/accounts/fireworks/routers/deepseek-v4p1-flash-1017-spot": { "cache_read_input_token_cost": 4.2e-09, "input_cost_per_token": 2.1e-07, @@ -82247,17 +82279,17 @@ "supports_native_streaming": true }, "fireworks_ai/accounts/fireworks/models/ultramarine-1-flash-us": { - "cache_read_input_token_cost": 4.5e-08, - "cache_read_input_token_cost_priority": 5.625e-08, - "input_cost_per_token": 2.25e-07, - "input_cost_per_token_priority": 2.8125e-07, + "cache_read_input_token_cost": 3e-08, + "cache_read_input_token_cost_priority": 3.75e-08, + "input_cost_per_token": 1.5e-07, + "input_cost_per_token_priority": 1.875e-07, "litellm_provider": "fireworks_ai", "max_input_tokens": 1048576, "max_output_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 7.5e-07, - "output_cost_per_token_priority": 9.375e-07, - "source": "https://docs.fireworks.ai/serverless/pricing", + "output_cost_per_token": 5e-07, + "output_cost_per_token_priority": 6.25e-07, + "source": "https://docs.fireworks.ai/serverless/pricing.md", "supports_function_calling": false, "supports_response_schema": true, "supports_tool_choice": false, diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index d8039c0d96e..157c79ca388 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -3,6 +3,8 @@ "cache_creation_input_audio_token_cost": 4e-07, "cache_read_input_audio_token_cost": 4e-07, "cache_read_input_token_cost": 4e-07, + "deprecation_date": "2027-06-25", + "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "input_cost_per_audio_token": 3.2e-05, "input_cost_per_image_token": 5e-06, "input_cost_per_token": 4e-06, @@ -42,6 +44,8 @@ "cache_creation_input_audio_token_cost": 3e-07, "cache_read_input_audio_token_cost": 3e-07, "cache_read_input_token_cost": 6e-08, + "deprecation_date": "2027-06-25", + "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "input_cost_per_audio_token": 1e-05, "input_cost_per_image_token": 8e-07, "input_cost_per_token": 6e-07, @@ -5348,7 +5352,7 @@ "azure/eu/gpt-5-mini-2025-08-07": { "cache_read_input_token_cost": 2.75e-08, "cache_read_input_token_cost_priority": 4.95e-08, - "deprecation_date": "2027-02-09", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/eu/gpt-5.4-mini", "input_cost_per_token": 2.75e-07, @@ -5392,7 +5396,7 @@ "cache_read_input_token_cost_batches": 1.375e-08 }, "azure/eu/gpt-5.1": { - "deprecation_date": "2027-05-15", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/eu/gpt-5.2", "cache_read_input_token_cost": 1.375e-07, @@ -5742,7 +5746,7 @@ "supports_vision": true }, "azure/global/gpt-5.1": { - "deprecation_date": "2027-05-15", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "cache_read_input_token_cost": 1.25e-07, "cache_read_input_token_cost_batches": 6.25e-08, @@ -6226,7 +6230,7 @@ "supports_vision": true }, "azure/gpt-4o-2024-05-13": { - "deprecation_date": "2026-12-09", + "deprecation_date": "2027-01-14", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/gpt-5.1", "input_cost_per_token": 5e-06, @@ -7044,7 +7048,7 @@ "azure/gpt-5.1-2025-11-13": { "cache_read_input_token_cost": 1.25e-07, "cache_read_input_token_cost_priority": 2.5e-07, - "deprecation_date": "2027-05-15", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/gpt-5.2", "input_cost_per_token": 1.25e-06, @@ -7287,7 +7291,7 @@ "supports_vision": true }, "azure/gpt-5-mini": { - "deprecation_date": "2027-02-09", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/gpt-5.4-mini", "cache_read_input_token_cost": 2.5e-08, @@ -7335,7 +7339,7 @@ "azure/gpt-5-mini-2025-08-07": { "cache_read_input_token_cost": 2.5e-08, "cache_read_input_token_cost_priority": 4.5e-08, - "deprecation_date": "2027-02-09", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/gpt-5.4-mini", "input_cost_per_token": 2.5e-07, @@ -7503,7 +7507,7 @@ "supports_vision": true }, "azure/gpt-5.1": { - "deprecation_date": "2027-05-15", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/gpt-5.2", "cache_read_input_token_cost": 1.25e-07, @@ -12384,7 +12388,7 @@ "azure/us/gpt-5-mini-2025-08-07": { "cache_read_input_token_cost": 2.75e-08, "cache_read_input_token_cost_priority": 4.95e-08, - "deprecation_date": "2027-02-09", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/us/gpt-5.4-mini", "input_cost_per_token": 2.75e-07, @@ -12471,7 +12475,7 @@ "cache_read_input_token_cost_batches": 2.75e-09 }, "azure/us/gpt-5.1": { - "deprecation_date": "2027-05-15", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/us/gpt-5.2", "cache_read_input_token_cost": 1.375e-07, @@ -31527,6 +31531,8 @@ "input_cost_per_audio_token_batches": 2.5e-07 }, "gemini/gemini-embedding-2-preview": { + "deprecation_date": "2026-08-10", + "deprecation_source": "https://ai.google.dev/gemini-api/docs/deprecations", "input_cost_per_audio_token": 6.5e-06, "input_cost_per_audio_token_batches": 3.25e-06, "input_cost_per_image_token": 4.5e-07, @@ -31542,6 +31548,7 @@ "output_vector_size": 3072, "rpm": 10000, "source": "https://ai.google.dev/gemini-api/docs/pricing", + "successor_model": "gemini/gemini-embedding-2", "supports_audio_input": true, "supports_embedding_image_input": true, "supports_multimodal": true, @@ -67587,11 +67594,12 @@ }, "xai/grok-imagine-image-quality": { "deprecation_date": "2026-11-02", - "deprecation_source": "https://docs.x.ai/docs/models", + "deprecation_source": "https://docs.x.ai/developers/migration/imagine-image-quality-nov-2", "input_cost_per_image": 0.05, "litellm_provider": "xai", "mode": "image_generation", "source": "https://docs.x.ai/docs/models", + "successor_model": "xai/grok-imagine-image-2.0", "supported_endpoints": [ "/v1/images/generations" ], @@ -67605,11 +67613,12 @@ }, "xai/grok-imagine-image-quality-20260403": { "deprecation_date": "2026-11-02", - "deprecation_source": "https://docs.x.ai/docs/models", + "deprecation_source": "https://docs.x.ai/developers/migration/imagine-image-quality-nov-2", "input_cost_per_image": 0.05, "litellm_provider": "xai", "mode": "image_generation", "source": "https://docs.x.ai/docs/models", + "successor_model": "xai/grok-imagine-image-2.0", "supported_endpoints": [ "/v1/images/generations" ], @@ -67623,11 +67632,12 @@ }, "xai/grok-imagine-image-quality-latest": { "deprecation_date": "2026-11-02", - "deprecation_source": "https://docs.x.ai/docs/models", + "deprecation_source": "https://docs.x.ai/developers/migration/imagine-image-quality-nov-2", "input_cost_per_image": 0.05, "litellm_provider": "xai", "mode": "image_generation", "source": "https://docs.x.ai/docs/models", + "successor_model": "xai/grok-imagine-image-2.0", "supported_endpoints": [ "/v1/images/generations" ], @@ -67641,11 +67651,12 @@ }, "xai/grok-imagine-image-pro": { "deprecation_date": "2026-05-15", - "deprecation_source": "https://docs.x.ai/docs/models", + "deprecation_source": "https://docs.x.ai/developers/migration/may-15-retirement", "input_cost_per_image": 0.05, "litellm_provider": "xai", "mode": "image_generation", "source": "https://docs.x.ai/docs/models", + "successor_model": "xai/grok-imagine-image-2.0", "supported_endpoints": [ "/v1/images/generations" ], @@ -67763,6 +67774,7 @@ }, "xai/grok-voice-transcribe-1.0": { "deprecation_date": "2026-10-02", + "deprecation_source": "https://docs.x.ai/docs/release-notes", "successor_model": "xai/grok-voice-transcribe-2.0", "input_cost_per_second": 2.778e-05, "litellm_provider": "xai", @@ -71226,7 +71238,7 @@ "source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'" }, "azure/eu/gpt-4o-2024-05-13": { - "deprecation_date": "2026-12-09", + "deprecation_date": "2027-01-14", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/eu/gpt-5.1", "input_cost_per_token": 5.5e-06, @@ -71273,7 +71285,7 @@ "source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'" }, "azure/eu/gpt-5-mini": { - "deprecation_date": "2027-02-09", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/eu/gpt-5.4-mini", "cache_read_input_token_cost": 2.75e-08, @@ -71846,7 +71858,7 @@ "source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'" }, "azure/us/gpt-4o-2024-05-13": { - "deprecation_date": "2026-12-09", + "deprecation_date": "2027-01-14", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/us/gpt-5.1", "input_cost_per_token": 5.5e-06, @@ -71893,7 +71905,7 @@ "source": "https://prices.azure.com/api/retail/prices?$filter=serviceName%20eq%20'Foundry%20Models'%20and%20armRegionName%20eq%20'eastus'%20and%20priceType%20eq%20'Consumption'" }, "azure/us/gpt-5-mini": { - "deprecation_date": "2027-02-09", + "deprecation_date": "2027-07-13", "deprecation_source": "https://learn.microsoft.com/en-us/azure/foundry/openai/concepts/model-retirement-schedule", "successor_model": "azure/us/gpt-5.4-mini", "cache_read_input_token_cost": 2.75e-08, @@ -74556,6 +74568,7 @@ "cache_read_input_token_cost": 2e-07, "cache_read_input_token_cost_above_200k_tokens": 4e-07, "deprecation_date": "2026-05-15", + "deprecation_source": "https://docs.x.ai/developers/migration/may-15-retirement", "input_cost_per_image_token": 1e-06, "input_cost_per_token": 1e-06, "input_cost_per_token_above_200k_tokens": 2e-06, @@ -74566,6 +74579,7 @@ "output_cost_per_token": 2e-06, "output_cost_per_token_above_200k_tokens": 4e-06, "source": "https://api.x.ai/v1/language-models", + "successor_model": "xai/grok-build-0.1", "supports_function_calling": true, "supports_prompt_caching": true, "supports_reasoning": true, @@ -77927,6 +77941,7 @@ "litellm_provider": "xai", "mode": "realtime", "source": "https://docs.x.ai/developers/pricing", + "successor_model": "xai/grok-voice-3.0", "supported_endpoints": [ "/v1/realtime" ], @@ -82153,6 +82168,23 @@ "supports_tool_choice": true, "supports_vision": true }, + "fireworks_ai/accounts/fireworks/routers/deepseek-v4p1-flash-1017-ultrafast": { + "cache_read_input_token_cost": 4.8e-08, + "input_cost_per_token": 2.4e-06, + "litellm_provider": "fireworks_ai", + "max_input_tokens": 1048576, + "max_output_tokens": 393216, + "mode": "chat", + "output_cost_per_token": 9.6e-06, + "source": "https://docs.fireworks.ai/serverless/pricing.md", + "supports_function_calling": true, + "supports_mid_conversation_system": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_tool_choice": true, + "supports_vision": true + }, "fireworks_ai/accounts/fireworks/routers/deepseek-v4p1-flash-1017-spot": { "cache_read_input_token_cost": 4.2e-09, "input_cost_per_token": 2.1e-07, @@ -82247,17 +82279,17 @@ "supports_native_streaming": true }, "fireworks_ai/accounts/fireworks/models/ultramarine-1-flash-us": { - "cache_read_input_token_cost": 4.5e-08, - "cache_read_input_token_cost_priority": 5.625e-08, - "input_cost_per_token": 2.25e-07, - "input_cost_per_token_priority": 2.8125e-07, + "cache_read_input_token_cost": 3e-08, + "cache_read_input_token_cost_priority": 3.75e-08, + "input_cost_per_token": 1.5e-07, + "input_cost_per_token_priority": 1.875e-07, "litellm_provider": "fireworks_ai", "max_input_tokens": 1048576, "max_output_tokens": 1048576, "mode": "chat", - "output_cost_per_token": 7.5e-07, - "output_cost_per_token_priority": 9.375e-07, - "source": "https://docs.fireworks.ai/serverless/pricing", + "output_cost_per_token": 5e-07, + "output_cost_per_token_priority": 6.25e-07, + "source": "https://docs.fireworks.ai/serverless/pricing.md", "supports_function_calling": false, "supports_response_schema": true, "supports_tool_choice": false,