diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 974f1bb0c8b..5a46a0891fc 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -2135,12 +2135,14 @@ "supports_output_config": true, "supports_parallel_tool_use_config": true, "prompt_cache_min_tokens": 512, - "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/index.json", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json", "token_count_model": { "model": "anthropic.claude-opus-5", "litellm_provider": "bedrock_mantle", "region": "us-east-1" - } + }, + "input_cost_per_token_batches": 2.5e-06, + "output_cost_per_token_batches": 1.25e-05 }, "anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -2221,12 +2223,14 @@ "supports_output_config": true, "supports_parallel_tool_use_config": true, "prompt_cache_min_tokens": 512, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json", "token_count_model": { "model": "global.anthropic.claude-opus-5", "litellm_provider": "bedrock_mantle", "region": "us-east-1" - } + }, + "input_cost_per_token_batches": 2.5e-06, + "output_cost_per_token_batches": 1.25e-05 }, "global.anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -2307,12 +2311,14 @@ "supports_output_config": true, "supports_parallel_tool_use_config": true, "prompt_cache_min_tokens": 512, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json", "token_count_model": { "model": "anthropic.claude-opus-5", "litellm_provider": "bedrock_mantle", "region": "us-east-1" - } + }, + "input_cost_per_token_batches": 2.75e-06, + "output_cost_per_token_batches": 1.375e-05 }, "us.anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -2393,12 +2399,14 @@ "supports_output_config": true, "supports_parallel_tool_use_config": true, "prompt_cache_min_tokens": 512, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/eu-west-1/index.json", "token_count_model": { "model": "anthropic.claude-opus-5", "litellm_provider": "bedrock_mantle", "region": "eu-west-1" - } + }, + "input_cost_per_token_batches": 2.75e-06, + "output_cost_per_token_batches": 1.375e-05 }, "eu.anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -2474,12 +2482,14 @@ "supports_output_config": true, "supports_parallel_tool_use_config": true, "prompt_cache_min_tokens": 512, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/ap-southeast-2/index.json", "token_count_model": { "model": "anthropic.claude-opus-5", "litellm_provider": "bedrock_mantle", "region": "ap-southeast-4" - } + }, + "input_cost_per_token_batches": 2.75e-06, + "output_cost_per_token_batches": 1.375e-05 }, "au.anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -2560,7 +2570,9 @@ "supports_output_config": true, "supports_parallel_tool_use_config": true, "prompt_cache_min_tokens": 512, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/ap-northeast-1/index.json", + "input_cost_per_token_batches": 2.75e-06, + "output_cost_per_token_batches": 1.375e-05 }, "jp.anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -3949,7 +3961,7 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "supports_max_reasoning_effort": true, + "supports_max_reasoning_effort": false, "prompt_cache_min_tokens": 512 }, "azure_ai/claude-fable-5-1": { @@ -3983,10 +3995,40 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "supports_max_reasoning_effort": true, + "supports_max_reasoning_effort": false, "prompt_cache_min_tokens": 512, "deprecation_date": "2027-12-05" }, + "azure_ai/claude-mythos-5-1": { + "input_cost_per_token": 1e-05, + "output_cost_per_token": 5e-05, + "cache_creation_input_token_cost": 1.25e-05, + "cache_creation_input_token_cost_above_1hr": 2e-05, + "cache_read_input_token_cost": 2.5e-07, + "litellm_provider": "azure_ai", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "mode": "chat", + "supports_adaptive_thinking": true, + "thinking_always_on": true, + "supports_mid_conversation_system": true, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_forced_tool_use": false, + "supports_function_calling": true, + "supports_native_structured_output": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_sampling_params": false, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_max_reasoning_effort": false, + "prompt_cache_min_tokens": 512, + "source": "https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/claude-models" + }, "azure_ai/claude-opus-5": { "deprecation_date": "2027-07-08", "supports_mid_conversation_system": true, @@ -22269,7 +22311,7 @@ }, "mode": "chat", "output_cost_per_token": 3.74997e-06, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "source": "https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models", "supported_modalities": [ "text", "image", @@ -22287,7 +22329,9 @@ "input_cost_per_token_priority": 1.35002e-06, "output_cost_per_token_priority": 6.75003e-06, "cache_read_input_token_cost_priority": 1.3503e-07, - "cache_creation_input_token_cost_priority": 1.35002e-06 + "cache_creation_input_token_cost_priority": 1.35002e-06, + "supports_audio_input": true, + "supports_video_input": true }, "databricks/databricks-gemini-3-7-flash": { "cache_creation_input_token_cost": 7.4998e-07, @@ -22301,7 +22345,7 @@ }, "mode": "chat", "output_cost_per_token": 3.74997e-06, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "source": "https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models", "supported_modalities": [ "text", "image", @@ -22319,7 +22363,9 @@ "input_cost_per_token_priority": 1.35002e-06, "output_cost_per_token_priority": 6.75003e-06, "cache_read_input_token_cost_priority": 1.3503e-07, - "cache_creation_input_token_cost_priority": 1.35002e-06 + "cache_creation_input_token_cost_priority": 1.35002e-06, + "supports_audio_input": true, + "supports_video_input": true }, "databricks/databricks-gemini-3-6-flash": { "cache_creation_input_token_cost": 7.4998e-07, @@ -22335,7 +22381,7 @@ "mode": "chat", "output_cost_per_token": 3.74997e-06, "output_dbu_cost_per_token": 5.3571e-05, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "source": "https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models", "supported_modalities": [ "text", "image", @@ -22354,7 +22400,9 @@ "input_cost_per_token_priority": 1.35002e-06, "output_cost_per_token_priority": 6.75003e-06, "cache_read_input_token_cost_priority": 1.3503e-07, - "cache_creation_input_token_cost_priority": 1.35002e-06 + "cache_creation_input_token_cost_priority": 1.35002e-06, + "supports_audio_input": true, + "supports_video_input": true }, "databricks/databricks-gemini-3-5-flash": { "cache_creation_input_token_cost": 1.87502e-06, @@ -22370,7 +22418,7 @@ "mode": "chat", "output_cost_per_token": 1.124998e-05, "output_dbu_cost_per_token": 0.000160714, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "source": "https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models", "supported_modalities": [ "text", "image", @@ -22388,7 +22436,9 @@ "input_cost_per_token_priority": 3.37498e-06, "output_cost_per_token_priority": 2.025002e-05, "cache_read_input_token_cost_priority": 3.3747e-07, - "cache_creation_input_token_cost_priority": 3.37498e-06 + "cache_creation_input_token_cost_priority": 3.37498e-06, + "supports_audio_input": true, + "supports_video_input": true }, "databricks/databricks-gemini-3-5-flash-lite": { "cache_creation_input_token_cost": 3.7499e-07, @@ -22404,7 +22454,7 @@ "mode": "chat", "output_cost_per_token": 3.12501e-06, "output_dbu_cost_per_token": 4.4643e-05, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "source": "https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models", "supported_modalities": [ "text", "image", @@ -22424,7 +22474,9 @@ "input_cost_per_token_priority": 6.7501e-07, "output_cost_per_token_priority": 5.62499e-06, "cache_read_input_token_cost_priority": 6.748e-08, - "cache_creation_input_token_cost_priority": 6.7501e-07 + "cache_creation_input_token_cost_priority": 6.7501e-07, + "supports_audio_input": true, + "supports_video_input": true }, "databricks/databricks-gemma-3-12b": { "cache_creation_input_token_cost": 1.5001e-07, @@ -24167,7 +24219,7 @@ "max_output_tokens": 163840, "input_cost_per_token": 5e-07, "output_cost_per_token": 2.15e-06, - "cache_read_input_token_cost": 4e-07, + "cache_read_input_token_cost": 3.5e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true, @@ -24241,7 +24293,7 @@ "max_output_tokens": 163840, "input_cost_per_token": 2.5e-07, "output_cost_per_token": 9.5e-07, - "cache_read_input_token_cost": 2.16e-07, + "cache_read_input_token_cost": 1.3e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true, @@ -29639,6 +29691,7 @@ "web_search_billing_unit": "per_query" }, "gemini-3.1-flash-lite-image": { + "deprecation_date": "2027-06-28", "cache_read_input_token_cost": 2.5e-08, "cache_read_input_token_cost_batches": 1.25e-08, "cache_read_input_token_cost_flex": 1.25e-08, @@ -30638,6 +30691,7 @@ "google_maps_grounding_cost_per_query": 0.014 }, "vertex_ai/gemini-3.8-flash": { + "deprecation_date": "2027-09-02", "cache_read_input_token_cost": 7.5e-08, "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, @@ -30744,6 +30798,7 @@ "supports_native_streaming": true }, "vertex_ai/gemini-3.8-live": { + "deprecation_date": "2027-09-25", "gemini_audio_only_live": true, "input_cost_per_audio_token": 3e-06, "input_cost_per_image_token": 1e-06, @@ -32706,13 +32761,13 @@ "input_cost_per_audio_token": 1.5e-06, "input_cost_per_token": 1.5e-06, "litellm_provider": "vertex_ai-language-models", - "max_input_tokens": 1048576, - "max_output_tokens": 65535, + "max_input_tokens": 131072, + "max_output_tokens": 57920, "mode": "chat", "output_cost_per_reasoning_token": 9e-06, "output_cost_per_token": 9e-06, "output_cost_per_video_token": 1.75e-05, - "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", + "source": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/omni-flash-preview", "supported_endpoints": [ "/v1/chat/completions" ], @@ -32730,7 +32785,8 @@ "supports_reasoning": true, "supports_system_messages": true, "supports_video_input": true, - "supports_vision": true + "supports_vision": true, + "deprecation_date": "2026-11-06" }, "gemini-3.5-flash": { "prompt_cache_min_tokens": 4096, @@ -32918,6 +32974,7 @@ "google_maps_grounding_cost_per_query": 0.014 }, "gemini-3.8-flash": { + "deprecation_date": "2027-09-02", "cache_read_input_token_cost": 7.5e-08, "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, @@ -44265,7 +44322,8 @@ "mode": "chat", "supports_function_calling": true, "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/google%2Fgemma-3-27b-it" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 4e-08 }, "nebius/meta-llama/Llama-3.3-70B-Instruct": { "max_input_tokens": 131072, @@ -44491,7 +44549,7 @@ "source": "https://tokenfactory.nebius.com/models/catalog/text2text/deepseek-ai%2FDeepSeek-V4-Flash" }, "nebius/deepseek-ai/DeepSeek-V4-Flash-0731": { - "max_input_tokens": 1024000, + "max_input_tokens": 1048576, "max_output_tokens": 1024000, "input_cost_per_token": 1.4e-07, "output_cost_per_token": 2.8e-07, @@ -44499,7 +44557,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/deepseek-ai%2FDeepSeek-V4-Flash-0731" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1e-08 }, "nebius/deepseek-ai/DeepSeek-V4-Pro": { "max_input_tokens": 1048576, @@ -44510,29 +44569,32 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/deepseek-ai%2FDeepSeek-V4-Pro" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1.4e-07 }, "nebius/deepseek-ai/DeepSeek-V4-Pro-0813": { "input_cost_per_token": 1.32e-06, "litellm_provider": "nebius", - "max_input_tokens": 979000, + "max_input_tokens": 1048576, "mode": "chat", "output_cost_per_token": 3.96e-06, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/deepseek-ai%2FDeepSeek-V4-Pro-0813", + "source": "https://tokenfactory.nebius.com/api/public/models_info", "supports_function_calling": true, - "supports_reasoning": true + "supports_reasoning": true, + "cache_read_input_token_cost": 4e-08 }, "nebius/deepseek-ai/DeepSeek-V4.1-Flash": { "input_cost_per_token": 3e-07, "litellm_provider": "nebius", - "max_input_tokens": 1048000, + "max_input_tokens": 1048576, "max_output_tokens": 384000, "mode": "chat", "output_cost_per_token": 1.2e-06, - "source": "https://tokenfactory.nebius.com/endpoints?modals=endpoint-details&model-id=deepseek-ai/DeepSeek-V4.1-Flash", + "source": "https://tokenfactory.nebius.com/api/public/models_info", "supports_function_calling": true, "supports_reasoning": true, - "supports_vision": true + "supports_vision": true, + "cache_read_input_token_cost": 1e-08 }, "nebius/MiniMaxAI/MiniMax-M2.5": { "max_input_tokens": 196608, @@ -44555,7 +44617,8 @@ "supports_function_calling": true, "supports_reasoning": true, "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/MiniMaxAI%2FMiniMax-M3" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 6e-08 }, "nebius/moonshotai/Kimi-K2.6": { "max_input_tokens": 262144, @@ -44567,7 +44630,8 @@ "supports_function_calling": true, "supports_reasoning": true, "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/image2text/moonshotai%2FKimi-K2.6" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1.6e-07 }, "nebius/moonshotai/Kimi-K2.7-Code": { "max_input_tokens": 262144, @@ -44579,10 +44643,11 @@ "supports_function_calling": true, "supports_reasoning": true, "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/moonshotai%2FKimi-K2.7-Code" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1.5e-07 }, "nebius/moonshotai/Kimi-K3": { - "max_input_tokens": 1024000, + "max_input_tokens": 1048576, "max_output_tokens": 1024000, "input_cost_per_token": 3e-06, "output_cost_per_token": 1.5e-05, @@ -44591,7 +44656,8 @@ "supports_function_calling": true, "supports_reasoning": true, "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/image2text/moonshotai%2FKimi-K3" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 3e-07 }, "nebius/NousResearch/Hermes-4-405B": { "max_input_tokens": 131072, @@ -44602,7 +44668,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/NousResearch%2FHermes-4-405B" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 2e-07 }, "nebius/NousResearch/Hermes-4-70B": { "max_input_tokens": 131072, @@ -44646,7 +44713,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2FNVIDIA-Nemotron-3-Nano-30B-A3B" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1.2e-08 }, "nebius/nvidia/Nemotron-3-Nano-Omni": { "max_input_tokens": 262144, @@ -44669,7 +44737,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2Fnemotron-3-super-120b-a12b" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 6e-08 }, "nebius/nvidia/Nemotron-3-Ultra-550b-a55b": { "max_input_tokens": 1048576, @@ -44680,7 +44749,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2FNemotron-3-Ultra-550b-a55b" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1.2e-07 }, "nebius/nvidia/Nemotron-3_5-Lightning": { "max_input_tokens": 1048576, @@ -44691,7 +44761,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2FNemotron-3_5-Lightning" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 3e-08 }, "nebius/openai/gpt-oss-120b": { "max_input_tokens": 131072, @@ -44702,7 +44773,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/openai%2Fgpt-oss-120b" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1e-07 }, "nebius/openbmb/MiniCPM-V-4_5": { "max_input_tokens": 32000, @@ -44712,7 +44784,8 @@ "litellm_provider": "nebius", "mode": "chat", "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/image2text/openbmb%2FMiniCPM-V-4_5" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1.3e-07 }, "nebius/Qwen/Qwen3-235B-A22B-Instruct-2507": { "max_input_tokens": 262144, @@ -44722,7 +44795,8 @@ "litellm_provider": "nebius", "mode": "chat", "supports_function_calling": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3-235B-A22B-Instruct-2507" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 5e-08 }, "nebius/Qwen/Qwen3-30B-A3B-Instruct-2507": { "max_input_tokens": 262144, @@ -44732,7 +44806,8 @@ "litellm_provider": "nebius", "mode": "chat", "supports_function_calling": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3-30B-A3B-Instruct-2507" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1e-07 }, "nebius/Qwen/Qwen3-Next-80B-A3B-Thinking": { "max_input_tokens": 128000, @@ -44754,7 +44829,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3.5-397B-A17B" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1.2e-07 }, "nebius/Qwen/Qwen3.8-27B": { "input_cost_per_token": 4.5e-07, @@ -44762,9 +44838,10 @@ "max_input_tokens": 262144, "mode": "chat", "output_cost_per_token": 3e-06, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3.8-27B", + "source": "https://tokenfactory.nebius.com/api/public/models_info", "supports_function_calling": true, - "supports_reasoning": true + "supports_reasoning": true, + "cache_read_input_token_cost": 1e-07 }, "nebius/zai-org/GLM-5.1": { "max_input_tokens": 202752, @@ -44775,7 +44852,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/zai-org%2FGLM-5.1" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 2.6e-07 }, "nebius/zai-org/GLM-5.2": { "max_input_tokens": 1048576, @@ -44786,21 +44864,23 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/zai-org%2FGLM-5.2" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1.4e-07 }, "nebius/zai-org/GLM-5.3": { "input_cost_per_token": 1.4e-06, "litellm_provider": "nebius", - "max_input_tokens": 1024000, + "max_input_tokens": 1048576, "max_output_tokens": 1024000, "mode": "chat", "output_cost_per_token": 4.4e-06, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/zai-org%2FGLM-5.3", + "source": "https://tokenfactory.nebius.com/api/public/models_info", "supports_function_calling": true, - "supports_reasoning": true + "supports_reasoning": true, + "cache_read_input_token_cost": 2.6e-07 }, "nebius/zai-org/GLM-5.3-Flash": { - "max_input_tokens": 1024000, + "max_input_tokens": 1048576, "max_output_tokens": 1024000, "input_cost_per_token": 1.5e-07, "output_cost_per_token": 5e-07, @@ -44808,8 +44888,9 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/zai-org%2FGLM-5.3-Flash", - "supports_vision": true + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "supports_vision": true, + "cache_read_input_token_cost": 3e-08 }, "nebius/BAAI/bge-en-icl": { "max_input_tokens": 32768, @@ -49813,7 +49894,7 @@ "mode": "chat", "output_cost_per_token": 3e-05, "prompt_cache_min_tokens": 512, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-gov-west-1/index.json", "supports_adaptive_thinking": true, "supports_assistant_prefill": false, "supports_computer_use": true, @@ -49835,7 +49916,9 @@ "model": "anthropic.claude-opus-5", "litellm_provider": "bedrock_mantle", "region": "us-gov-west-1" - } + }, + "input_cost_per_token_batches": 3e-06, + "output_cost_per_token_batches": 1.5e-05 }, "us-gov.anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -52922,6 +53005,7 @@ }, "vertex_ai/gemini-3.1-flash-lite-image": { + "deprecation_date": "2027-06-28", "cache_read_input_token_cost": 2.5e-08, "cache_read_input_token_cost_batches": 1.25e-08, "cache_read_input_token_cost_flex": 1.25e-08, @@ -54150,7 +54234,7 @@ "voyage/voyage-lite-01": { "input_cost_per_token": 1e-07, "litellm_provider": "voyage", - "max_input_tokens": 4096, + "max_input_tokens": 4000, "mode": "embedding", "output_cost_per_token": 0.0 }, @@ -64809,7 +64893,7 @@ "supports_vision": false }, "novita/deepseek/deepseek-v4-pro-0813": { - "cache_read_input_token_cost": 1.3200000000000002e-07, + "cache_read_input_token_cost": 4.4e-08, "input_cost_per_token": 1.32e-06, "litellm_provider": "novita", "max_input_tokens": 1048576, @@ -64921,13 +65005,13 @@ "supports_vision": false }, "novita/mindai/macaron-v1-venti": { - "cache_read_input_token_cost": 3e-07, - "input_cost_per_token": 1.5e-06, + "cache_read_input_token_cost": 1.65e-07, + "input_cost_per_token": 8.25e-07, "litellm_provider": "novita", "max_input_tokens": 1048576, "max_output_tokens": 131072, "mode": "chat", - "output_cost_per_token": 4.5e-06, + "output_cost_per_token": 2.475e-06, "source": "https://novita.ai/pricing", "supports_function_calling": true, "supports_prompt_caching": true, @@ -65030,13 +65114,13 @@ "supports_vision": false }, "novita/mindai/macaron-v1-tall": { - "cache_read_input_token_cost": 8e-08, - "input_cost_per_token": 4.5000000000000003e-07, + "cache_read_input_token_cost": 4.4e-08, + "input_cost_per_token": 2.475e-07, "litellm_provider": "novita", "max_input_tokens": 262144, "max_output_tokens": 32768, "mode": "chat", - "output_cost_per_token": 2.6e-06, + "output_cost_per_token": 1.43e-06, "source": "https://novita.ai/pricing", "supports_function_calling": true, "supports_prompt_caching": true, @@ -66461,9 +66545,9 @@ "deepinfra/deepseek-ai/DeepSeek-V4-Flash-0731": { "max_output_tokens": 1048576, "max_input_tokens": 1048576, - "input_cost_per_token": 8e-08, + "input_cost_per_token": 6e-08, "output_cost_per_token": 1.8e-07, - "cache_read_input_token_cost": 1.6e-08, + "cache_read_input_token_cost": 1.5e-08, "supports_prompt_caching": true, "litellm_provider": "deepinfra", "mode": "chat", @@ -66842,8 +66926,8 @@ "deepinfra/google/gemma-4-31B-it": { "max_output_tokens": 262144, "max_input_tokens": 262144, - "input_cost_per_token": 1.3e-07, - "output_cost_per_token": 3.8e-07, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_function_calling": true, @@ -68554,7 +68638,10 @@ "model": "anthropic.claude-opus-5", "litellm_provider": "bedrock_mantle", "region": "us-gov-west-1" - } + }, + "input_cost_per_token_batches": 3e-06, + "output_cost_per_token_batches": 1.5e-05, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-gov-west-1/index.json" }, "bedrock/us-gov-west-1/anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -68829,7 +68916,10 @@ "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, - "supports_xhigh_reasoning_effort": true + "supports_xhigh_reasoning_effort": true, + "input_cost_per_token_batches": 3e-06, + "output_cost_per_token_batches": 1.5e-05, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-gov-east-1/index.json" }, "bedrock/us-gov-east-1/anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -69852,6 +69942,14 @@ "cache_read_input_token_cost": 3e-08, "source": "https://docs.perplexity.ai/docs/agent-api/models" }, + "perplexity/google/gemini-4-argon": { + "litellm_provider": "perplexity", + "mode": "responses", + "input_cost_per_token": 2e-06, + "output_cost_per_token": 1e-05, + "cache_read_input_token_cost": 1e-07, + "source": "https://docs.perplexity.ai/docs/agent-api/models" + }, "perplexity/google/gemini-3.6-flash": { "litellm_provider": "perplexity", "mode": "responses", @@ -69979,6 +70077,14 @@ "cache_read_input_token_cost": 1.15e-09, "source": "https://docs.perplexity.ai/docs/agent-api/models" }, + "perplexity/perplexity/nemotron-3.5-super-120b-a12b": { + "litellm_provider": "perplexity", + "mode": "responses", + "input_cost_per_token": 2e-07, + "output_cost_per_token": 5e-07, + "cache_read_input_token_cost": 4e-08, + "source": "https://docs.perplexity.ai/docs/agent-api/models" + }, "perplexity/perplexity/nemotron-3-ultra-550b-a55b": { "litellm_provider": "perplexity", "mode": "responses", @@ -70104,12 +70210,12 @@ "input_cost_per_token": 1.5e-06, "litellm_provider": "vertex_ai", "max_output_tokens": 57920, - "max_input_tokens": 57920, + "max_input_tokens": 131072, "mode": "chat", "output_cost_per_reasoning_token": 9e-06, "output_cost_per_token": 9e-06, "output_cost_per_video_token": 1.75e-05, - "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", + "source": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/omni-1-1-flash", "supported_endpoints": [ "/v1beta/interactions" ], @@ -74505,12 +74611,14 @@ "input_cost_per_token": 5.5e-06, "cache_read_input_token_cost": 5.5e-07, "output_cost_per_token": 2.75e-05, - "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/index.json", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/ap-northeast-1/index.json", "token_count_model": { "model": "anthropic.claude-opus-5", "litellm_provider": "bedrock_mantle", "region": "ap-southeast-4" - } + }, + "input_cost_per_token_batches": 2.75e-06, + "output_cost_per_token_batches": 1.375e-05 }, "apac.anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -75107,7 +75215,7 @@ "supports_function_calling": true, "input_cost_per_token": 2e-06, "output_cost_per_token": 1e-05, - "cache_read_input_token_cost": 2e-07, + "cache_read_input_token_cost": 1e-07, "source": "https://docs.perplexity.ai/docs/agent-api/models" }, "perplexity/anthropic/claude-fable-5-1": { @@ -75155,6 +75263,9 @@ "output_cost_per_token": 5e-05, "cache_read_input_token_cost": 1e-06, "cache_creation_input_token_cost": 1.25e-05, + "input_cost_per_token_above_272k_tokens": 2e-05, + "output_cost_per_token_above_272k_tokens": 7.5e-05, + "cache_read_input_token_cost_above_272k_tokens": 2e-06, "source": "https://docs.perplexity.ai/docs/agent-api/models" }, "perplexity/openai/gpt-6-luna": { @@ -76622,6 +76733,9 @@ "input_cost_per_token": 1e-07, "output_cost_per_token": 5e-07, "cache_read_input_token_cost": 1e-08, + "input_cost_per_token_above_100k_tokens": 5e-07, + "output_cost_per_token_above_100k_tokens": 2.5e-06, + "cache_read_input_token_cost_above_100k_tokens": 5e-08, "source": "https://docs.perplexity.ai/docs/agent-api/models" }, "us-gov.anthropic.claude-haiku-5-5": { @@ -77096,6 +77210,7 @@ "web_search_billing_unit": "per_query" }, "vertex_ai/gemini-4-argon": { + "deprecation_date": "2027-10-13", "cache_read_input_token_cost": 1e-07, "cache_read_input_token_cost_batches": 5e-08, "cache_read_input_token_cost_flex": 5e-08, @@ -77504,9 +77619,10 @@ "max_output_tokens": 1048576, "mode": "chat", "output_cost_per_token": 1e-06, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/reflection%2Fbeam", + "source": "https://tokenfactory.nebius.com/api/public/models_info", "supports_function_calling": true, - "supports_reasoning": true + "supports_reasoning": true, + "cache_read_input_token_cost": 5e-08 }, "baseten/deepseek-ai/DeepSeek-V4.1-Flash-1017": { "cache_read_input_token_cost": 7e-09, @@ -81777,5 +81893,55 @@ "supports_response_schema": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true + }, + "nebius/deepseek-ai/DeepSeek-V4.1-Flash-1017": { + "max_input_tokens": 1048576, + "input_cost_per_token": 3e-07, + "cache_read_input_token_cost": 6e-09, + "output_cost_per_token": 1.2e-06, + "litellm_provider": "nebius", + "mode": "chat", + "supports_vision": true, + "supports_reasoning": true, + "supports_function_calling": true, + "source": "https://tokenfactory.nebius.com/api/public/models_info" + }, + "nebius/meta-models/muse-spark-open": { + "max_input_tokens": 1048576, + "input_cost_per_token": 1.25e-06, + "cache_read_input_token_cost": 1.5e-07, + "output_cost_per_token": 4.25e-06, + "litellm_provider": "nebius", + "mode": "chat", + "supports_vision": true, + "supports_video_input": true, + "supports_reasoning": true, + "supports_function_calling": true, + "source": "https://tokenfactory.nebius.com/api/public/models_info" + }, + "nebius/MiniMaxAI/MiniMax-M3.1-Flash": { + "max_input_tokens": 1048576, + "input_cost_per_token": 1.5e-07, + "cache_read_input_token_cost": 1.5e-08, + "output_cost_per_token": 6e-07, + "litellm_provider": "nebius", + "mode": "chat", + "supports_vision": true, + "supports_reasoning": true, + "supports_function_calling": true, + "source": "https://tokenfactory.nebius.com/api/public/models_info" + }, + "nebius/nvidia/Nemotron-3.5-Super-VL": { + "max_input_tokens": 1048576, + "input_cost_per_token": 3e-07, + "cache_read_input_token_cost": 6e-08, + "output_cost_per_token": 9e-07, + "litellm_provider": "nebius", + "mode": "chat", + "supports_vision": true, + "supports_video_input": true, + "supports_reasoning": true, + "supports_function_calling": true, + "source": "https://tokenfactory.nebius.com/api/public/models_info" } } diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 974f1bb0c8b..5a46a0891fc 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -2135,12 +2135,14 @@ "supports_output_config": true, "supports_parallel_tool_use_config": true, "prompt_cache_min_tokens": 512, - "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/index.json", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json", "token_count_model": { "model": "anthropic.claude-opus-5", "litellm_provider": "bedrock_mantle", "region": "us-east-1" - } + }, + "input_cost_per_token_batches": 2.5e-06, + "output_cost_per_token_batches": 1.25e-05 }, "anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -2221,12 +2223,14 @@ "supports_output_config": true, "supports_parallel_tool_use_config": true, "prompt_cache_min_tokens": 512, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json", "token_count_model": { "model": "global.anthropic.claude-opus-5", "litellm_provider": "bedrock_mantle", "region": "us-east-1" - } + }, + "input_cost_per_token_batches": 2.5e-06, + "output_cost_per_token_batches": 1.25e-05 }, "global.anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -2307,12 +2311,14 @@ "supports_output_config": true, "supports_parallel_tool_use_config": true, "prompt_cache_min_tokens": 512, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json", "token_count_model": { "model": "anthropic.claude-opus-5", "litellm_provider": "bedrock_mantle", "region": "us-east-1" - } + }, + "input_cost_per_token_batches": 2.75e-06, + "output_cost_per_token_batches": 1.375e-05 }, "us.anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -2393,12 +2399,14 @@ "supports_output_config": true, "supports_parallel_tool_use_config": true, "prompt_cache_min_tokens": 512, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/eu-west-1/index.json", "token_count_model": { "model": "anthropic.claude-opus-5", "litellm_provider": "bedrock_mantle", "region": "eu-west-1" - } + }, + "input_cost_per_token_batches": 2.75e-06, + "output_cost_per_token_batches": 1.375e-05 }, "eu.anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -2474,12 +2482,14 @@ "supports_output_config": true, "supports_parallel_tool_use_config": true, "prompt_cache_min_tokens": 512, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/ap-southeast-2/index.json", "token_count_model": { "model": "anthropic.claude-opus-5", "litellm_provider": "bedrock_mantle", "region": "ap-southeast-4" - } + }, + "input_cost_per_token_batches": 2.75e-06, + "output_cost_per_token_batches": 1.375e-05 }, "au.anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -2560,7 +2570,9 @@ "supports_output_config": true, "supports_parallel_tool_use_config": true, "prompt_cache_min_tokens": 512, - "source": "https://aws.amazon.com/bedrock/pricing/" + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/ap-northeast-1/index.json", + "input_cost_per_token_batches": 2.75e-06, + "output_cost_per_token_batches": 1.375e-05 }, "jp.anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -3949,7 +3961,7 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "supports_max_reasoning_effort": true, + "supports_max_reasoning_effort": false, "prompt_cache_min_tokens": 512 }, "azure_ai/claude-fable-5-1": { @@ -3983,10 +3995,40 @@ "supports_tool_choice": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true, - "supports_max_reasoning_effort": true, + "supports_max_reasoning_effort": false, "prompt_cache_min_tokens": 512, "deprecation_date": "2027-12-05" }, + "azure_ai/claude-mythos-5-1": { + "input_cost_per_token": 1e-05, + "output_cost_per_token": 5e-05, + "cache_creation_input_token_cost": 1.25e-05, + "cache_creation_input_token_cost_above_1hr": 2e-05, + "cache_read_input_token_cost": 2.5e-07, + "litellm_provider": "azure_ai", + "max_input_tokens": 1000000, + "max_output_tokens": 128000, + "mode": "chat", + "supports_adaptive_thinking": true, + "thinking_always_on": true, + "supports_mid_conversation_system": true, + "supports_assistant_prefill": false, + "supports_computer_use": true, + "supports_forced_tool_use": false, + "supports_function_calling": true, + "supports_native_structured_output": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_sampling_params": false, + "supports_tool_choice": true, + "supports_vision": true, + "supports_xhigh_reasoning_effort": true, + "supports_max_reasoning_effort": false, + "prompt_cache_min_tokens": 512, + "source": "https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/claude-models" + }, "azure_ai/claude-opus-5": { "deprecation_date": "2027-07-08", "supports_mid_conversation_system": true, @@ -22269,7 +22311,7 @@ }, "mode": "chat", "output_cost_per_token": 3.74997e-06, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "source": "https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models", "supported_modalities": [ "text", "image", @@ -22287,7 +22329,9 @@ "input_cost_per_token_priority": 1.35002e-06, "output_cost_per_token_priority": 6.75003e-06, "cache_read_input_token_cost_priority": 1.3503e-07, - "cache_creation_input_token_cost_priority": 1.35002e-06 + "cache_creation_input_token_cost_priority": 1.35002e-06, + "supports_audio_input": true, + "supports_video_input": true }, "databricks/databricks-gemini-3-7-flash": { "cache_creation_input_token_cost": 7.4998e-07, @@ -22301,7 +22345,7 @@ }, "mode": "chat", "output_cost_per_token": 3.74997e-06, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "source": "https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models", "supported_modalities": [ "text", "image", @@ -22319,7 +22363,9 @@ "input_cost_per_token_priority": 1.35002e-06, "output_cost_per_token_priority": 6.75003e-06, "cache_read_input_token_cost_priority": 1.3503e-07, - "cache_creation_input_token_cost_priority": 1.35002e-06 + "cache_creation_input_token_cost_priority": 1.35002e-06, + "supports_audio_input": true, + "supports_video_input": true }, "databricks/databricks-gemini-3-6-flash": { "cache_creation_input_token_cost": 7.4998e-07, @@ -22335,7 +22381,7 @@ "mode": "chat", "output_cost_per_token": 3.74997e-06, "output_dbu_cost_per_token": 5.3571e-05, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "source": "https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models", "supported_modalities": [ "text", "image", @@ -22354,7 +22400,9 @@ "input_cost_per_token_priority": 1.35002e-06, "output_cost_per_token_priority": 6.75003e-06, "cache_read_input_token_cost_priority": 1.3503e-07, - "cache_creation_input_token_cost_priority": 1.35002e-06 + "cache_creation_input_token_cost_priority": 1.35002e-06, + "supports_audio_input": true, + "supports_video_input": true }, "databricks/databricks-gemini-3-5-flash": { "cache_creation_input_token_cost": 1.87502e-06, @@ -22370,7 +22418,7 @@ "mode": "chat", "output_cost_per_token": 1.124998e-05, "output_dbu_cost_per_token": 0.000160714, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "source": "https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models", "supported_modalities": [ "text", "image", @@ -22388,7 +22436,9 @@ "input_cost_per_token_priority": 3.37498e-06, "output_cost_per_token_priority": 2.025002e-05, "cache_read_input_token_cost_priority": 3.3747e-07, - "cache_creation_input_token_cost_priority": 3.37498e-06 + "cache_creation_input_token_cost_priority": 3.37498e-06, + "supports_audio_input": true, + "supports_video_input": true }, "databricks/databricks-gemini-3-5-flash-lite": { "cache_creation_input_token_cost": 3.7499e-07, @@ -22404,7 +22454,7 @@ "mode": "chat", "output_cost_per_token": 3.12501e-06, "output_dbu_cost_per_token": 4.4643e-05, - "source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving", + "source": "https://docs.databricks.com/aws/en/machine-learning/foundation-model-apis/supported-models", "supported_modalities": [ "text", "image", @@ -22424,7 +22474,9 @@ "input_cost_per_token_priority": 6.7501e-07, "output_cost_per_token_priority": 5.62499e-06, "cache_read_input_token_cost_priority": 6.748e-08, - "cache_creation_input_token_cost_priority": 6.7501e-07 + "cache_creation_input_token_cost_priority": 6.7501e-07, + "supports_audio_input": true, + "supports_video_input": true }, "databricks/databricks-gemma-3-12b": { "cache_creation_input_token_cost": 1.5001e-07, @@ -24167,7 +24219,7 @@ "max_output_tokens": 163840, "input_cost_per_token": 5e-07, "output_cost_per_token": 2.15e-06, - "cache_read_input_token_cost": 4e-07, + "cache_read_input_token_cost": 3.5e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true, @@ -24241,7 +24293,7 @@ "max_output_tokens": 163840, "input_cost_per_token": 2.5e-07, "output_cost_per_token": 9.5e-07, - "cache_read_input_token_cost": 2.16e-07, + "cache_read_input_token_cost": 1.3e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true, @@ -29639,6 +29691,7 @@ "web_search_billing_unit": "per_query" }, "gemini-3.1-flash-lite-image": { + "deprecation_date": "2027-06-28", "cache_read_input_token_cost": 2.5e-08, "cache_read_input_token_cost_batches": 1.25e-08, "cache_read_input_token_cost_flex": 1.25e-08, @@ -30638,6 +30691,7 @@ "google_maps_grounding_cost_per_query": 0.014 }, "vertex_ai/gemini-3.8-flash": { + "deprecation_date": "2027-09-02", "cache_read_input_token_cost": 7.5e-08, "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, @@ -30744,6 +30798,7 @@ "supports_native_streaming": true }, "vertex_ai/gemini-3.8-live": { + "deprecation_date": "2027-09-25", "gemini_audio_only_live": true, "input_cost_per_audio_token": 3e-06, "input_cost_per_image_token": 1e-06, @@ -32706,13 +32761,13 @@ "input_cost_per_audio_token": 1.5e-06, "input_cost_per_token": 1.5e-06, "litellm_provider": "vertex_ai-language-models", - "max_input_tokens": 1048576, - "max_output_tokens": 65535, + "max_input_tokens": 131072, + "max_output_tokens": 57920, "mode": "chat", "output_cost_per_reasoning_token": 9e-06, "output_cost_per_token": 9e-06, "output_cost_per_video_token": 1.75e-05, - "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", + "source": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/omni-flash-preview", "supported_endpoints": [ "/v1/chat/completions" ], @@ -32730,7 +32785,8 @@ "supports_reasoning": true, "supports_system_messages": true, "supports_video_input": true, - "supports_vision": true + "supports_vision": true, + "deprecation_date": "2026-11-06" }, "gemini-3.5-flash": { "prompt_cache_min_tokens": 4096, @@ -32918,6 +32974,7 @@ "google_maps_grounding_cost_per_query": 0.014 }, "gemini-3.8-flash": { + "deprecation_date": "2027-09-02", "cache_read_input_token_cost": 7.5e-08, "cache_read_input_token_cost_batches": 3.75e-08, "cache_read_input_token_cost_flex": 3.75e-08, @@ -44265,7 +44322,8 @@ "mode": "chat", "supports_function_calling": true, "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/google%2Fgemma-3-27b-it" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 4e-08 }, "nebius/meta-llama/Llama-3.3-70B-Instruct": { "max_input_tokens": 131072, @@ -44491,7 +44549,7 @@ "source": "https://tokenfactory.nebius.com/models/catalog/text2text/deepseek-ai%2FDeepSeek-V4-Flash" }, "nebius/deepseek-ai/DeepSeek-V4-Flash-0731": { - "max_input_tokens": 1024000, + "max_input_tokens": 1048576, "max_output_tokens": 1024000, "input_cost_per_token": 1.4e-07, "output_cost_per_token": 2.8e-07, @@ -44499,7 +44557,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/deepseek-ai%2FDeepSeek-V4-Flash-0731" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1e-08 }, "nebius/deepseek-ai/DeepSeek-V4-Pro": { "max_input_tokens": 1048576, @@ -44510,29 +44569,32 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/deepseek-ai%2FDeepSeek-V4-Pro" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1.4e-07 }, "nebius/deepseek-ai/DeepSeek-V4-Pro-0813": { "input_cost_per_token": 1.32e-06, "litellm_provider": "nebius", - "max_input_tokens": 979000, + "max_input_tokens": 1048576, "mode": "chat", "output_cost_per_token": 3.96e-06, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/deepseek-ai%2FDeepSeek-V4-Pro-0813", + "source": "https://tokenfactory.nebius.com/api/public/models_info", "supports_function_calling": true, - "supports_reasoning": true + "supports_reasoning": true, + "cache_read_input_token_cost": 4e-08 }, "nebius/deepseek-ai/DeepSeek-V4.1-Flash": { "input_cost_per_token": 3e-07, "litellm_provider": "nebius", - "max_input_tokens": 1048000, + "max_input_tokens": 1048576, "max_output_tokens": 384000, "mode": "chat", "output_cost_per_token": 1.2e-06, - "source": "https://tokenfactory.nebius.com/endpoints?modals=endpoint-details&model-id=deepseek-ai/DeepSeek-V4.1-Flash", + "source": "https://tokenfactory.nebius.com/api/public/models_info", "supports_function_calling": true, "supports_reasoning": true, - "supports_vision": true + "supports_vision": true, + "cache_read_input_token_cost": 1e-08 }, "nebius/MiniMaxAI/MiniMax-M2.5": { "max_input_tokens": 196608, @@ -44555,7 +44617,8 @@ "supports_function_calling": true, "supports_reasoning": true, "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/MiniMaxAI%2FMiniMax-M3" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 6e-08 }, "nebius/moonshotai/Kimi-K2.6": { "max_input_tokens": 262144, @@ -44567,7 +44630,8 @@ "supports_function_calling": true, "supports_reasoning": true, "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/image2text/moonshotai%2FKimi-K2.6" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1.6e-07 }, "nebius/moonshotai/Kimi-K2.7-Code": { "max_input_tokens": 262144, @@ -44579,10 +44643,11 @@ "supports_function_calling": true, "supports_reasoning": true, "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/moonshotai%2FKimi-K2.7-Code" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1.5e-07 }, "nebius/moonshotai/Kimi-K3": { - "max_input_tokens": 1024000, + "max_input_tokens": 1048576, "max_output_tokens": 1024000, "input_cost_per_token": 3e-06, "output_cost_per_token": 1.5e-05, @@ -44591,7 +44656,8 @@ "supports_function_calling": true, "supports_reasoning": true, "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/image2text/moonshotai%2FKimi-K3" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 3e-07 }, "nebius/NousResearch/Hermes-4-405B": { "max_input_tokens": 131072, @@ -44602,7 +44668,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/NousResearch%2FHermes-4-405B" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 2e-07 }, "nebius/NousResearch/Hermes-4-70B": { "max_input_tokens": 131072, @@ -44646,7 +44713,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2FNVIDIA-Nemotron-3-Nano-30B-A3B" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1.2e-08 }, "nebius/nvidia/Nemotron-3-Nano-Omni": { "max_input_tokens": 262144, @@ -44669,7 +44737,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2Fnemotron-3-super-120b-a12b" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 6e-08 }, "nebius/nvidia/Nemotron-3-Ultra-550b-a55b": { "max_input_tokens": 1048576, @@ -44680,7 +44749,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2FNemotron-3-Ultra-550b-a55b" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1.2e-07 }, "nebius/nvidia/Nemotron-3_5-Lightning": { "max_input_tokens": 1048576, @@ -44691,7 +44761,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/nvidia%2FNemotron-3_5-Lightning" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 3e-08 }, "nebius/openai/gpt-oss-120b": { "max_input_tokens": 131072, @@ -44702,7 +44773,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/openai%2Fgpt-oss-120b" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1e-07 }, "nebius/openbmb/MiniCPM-V-4_5": { "max_input_tokens": 32000, @@ -44712,7 +44784,8 @@ "litellm_provider": "nebius", "mode": "chat", "supports_vision": true, - "source": "https://tokenfactory.nebius.com/models/catalog/image2text/openbmb%2FMiniCPM-V-4_5" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1.3e-07 }, "nebius/Qwen/Qwen3-235B-A22B-Instruct-2507": { "max_input_tokens": 262144, @@ -44722,7 +44795,8 @@ "litellm_provider": "nebius", "mode": "chat", "supports_function_calling": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3-235B-A22B-Instruct-2507" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 5e-08 }, "nebius/Qwen/Qwen3-30B-A3B-Instruct-2507": { "max_input_tokens": 262144, @@ -44732,7 +44806,8 @@ "litellm_provider": "nebius", "mode": "chat", "supports_function_calling": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3-30B-A3B-Instruct-2507" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1e-07 }, "nebius/Qwen/Qwen3-Next-80B-A3B-Thinking": { "max_input_tokens": 128000, @@ -44754,7 +44829,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3.5-397B-A17B" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1.2e-07 }, "nebius/Qwen/Qwen3.8-27B": { "input_cost_per_token": 4.5e-07, @@ -44762,9 +44838,10 @@ "max_input_tokens": 262144, "mode": "chat", "output_cost_per_token": 3e-06, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/Qwen%2FQwen3.8-27B", + "source": "https://tokenfactory.nebius.com/api/public/models_info", "supports_function_calling": true, - "supports_reasoning": true + "supports_reasoning": true, + "cache_read_input_token_cost": 1e-07 }, "nebius/zai-org/GLM-5.1": { "max_input_tokens": 202752, @@ -44775,7 +44852,8 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/zai-org%2FGLM-5.1" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 2.6e-07 }, "nebius/zai-org/GLM-5.2": { "max_input_tokens": 1048576, @@ -44786,21 +44864,23 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/zai-org%2FGLM-5.2" + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "cache_read_input_token_cost": 1.4e-07 }, "nebius/zai-org/GLM-5.3": { "input_cost_per_token": 1.4e-06, "litellm_provider": "nebius", - "max_input_tokens": 1024000, + "max_input_tokens": 1048576, "max_output_tokens": 1024000, "mode": "chat", "output_cost_per_token": 4.4e-06, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/zai-org%2FGLM-5.3", + "source": "https://tokenfactory.nebius.com/api/public/models_info", "supports_function_calling": true, - "supports_reasoning": true + "supports_reasoning": true, + "cache_read_input_token_cost": 2.6e-07 }, "nebius/zai-org/GLM-5.3-Flash": { - "max_input_tokens": 1024000, + "max_input_tokens": 1048576, "max_output_tokens": 1024000, "input_cost_per_token": 1.5e-07, "output_cost_per_token": 5e-07, @@ -44808,8 +44888,9 @@ "mode": "chat", "supports_function_calling": true, "supports_reasoning": true, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/zai-org%2FGLM-5.3-Flash", - "supports_vision": true + "source": "https://tokenfactory.nebius.com/api/public/models_info", + "supports_vision": true, + "cache_read_input_token_cost": 3e-08 }, "nebius/BAAI/bge-en-icl": { "max_input_tokens": 32768, @@ -49813,7 +49894,7 @@ "mode": "chat", "output_cost_per_token": 3e-05, "prompt_cache_min_tokens": 512, - "source": "https://aws.amazon.com/bedrock/pricing/", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-gov-west-1/index.json", "supports_adaptive_thinking": true, "supports_assistant_prefill": false, "supports_computer_use": true, @@ -49835,7 +49916,9 @@ "model": "anthropic.claude-opus-5", "litellm_provider": "bedrock_mantle", "region": "us-gov-west-1" - } + }, + "input_cost_per_token_batches": 3e-06, + "output_cost_per_token_batches": 1.5e-05 }, "us-gov.anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -52922,6 +53005,7 @@ }, "vertex_ai/gemini-3.1-flash-lite-image": { + "deprecation_date": "2027-06-28", "cache_read_input_token_cost": 2.5e-08, "cache_read_input_token_cost_batches": 1.25e-08, "cache_read_input_token_cost_flex": 1.25e-08, @@ -54150,7 +54234,7 @@ "voyage/voyage-lite-01": { "input_cost_per_token": 1e-07, "litellm_provider": "voyage", - "max_input_tokens": 4096, + "max_input_tokens": 4000, "mode": "embedding", "output_cost_per_token": 0.0 }, @@ -64809,7 +64893,7 @@ "supports_vision": false }, "novita/deepseek/deepseek-v4-pro-0813": { - "cache_read_input_token_cost": 1.3200000000000002e-07, + "cache_read_input_token_cost": 4.4e-08, "input_cost_per_token": 1.32e-06, "litellm_provider": "novita", "max_input_tokens": 1048576, @@ -64921,13 +65005,13 @@ "supports_vision": false }, "novita/mindai/macaron-v1-venti": { - "cache_read_input_token_cost": 3e-07, - "input_cost_per_token": 1.5e-06, + "cache_read_input_token_cost": 1.65e-07, + "input_cost_per_token": 8.25e-07, "litellm_provider": "novita", "max_input_tokens": 1048576, "max_output_tokens": 131072, "mode": "chat", - "output_cost_per_token": 4.5e-06, + "output_cost_per_token": 2.475e-06, "source": "https://novita.ai/pricing", "supports_function_calling": true, "supports_prompt_caching": true, @@ -65030,13 +65114,13 @@ "supports_vision": false }, "novita/mindai/macaron-v1-tall": { - "cache_read_input_token_cost": 8e-08, - "input_cost_per_token": 4.5000000000000003e-07, + "cache_read_input_token_cost": 4.4e-08, + "input_cost_per_token": 2.475e-07, "litellm_provider": "novita", "max_input_tokens": 262144, "max_output_tokens": 32768, "mode": "chat", - "output_cost_per_token": 2.6e-06, + "output_cost_per_token": 1.43e-06, "source": "https://novita.ai/pricing", "supports_function_calling": true, "supports_prompt_caching": true, @@ -66461,9 +66545,9 @@ "deepinfra/deepseek-ai/DeepSeek-V4-Flash-0731": { "max_output_tokens": 1048576, "max_input_tokens": 1048576, - "input_cost_per_token": 8e-08, + "input_cost_per_token": 6e-08, "output_cost_per_token": 1.8e-07, - "cache_read_input_token_cost": 1.6e-08, + "cache_read_input_token_cost": 1.5e-08, "supports_prompt_caching": true, "litellm_provider": "deepinfra", "mode": "chat", @@ -66842,8 +66926,8 @@ "deepinfra/google/gemma-4-31B-it": { "max_output_tokens": 262144, "max_input_tokens": 262144, - "input_cost_per_token": 1.3e-07, - "output_cost_per_token": 3.8e-07, + "input_cost_per_token": 2e-07, + "output_cost_per_token": 4e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_function_calling": true, @@ -68554,7 +68638,10 @@ "model": "anthropic.claude-opus-5", "litellm_provider": "bedrock_mantle", "region": "us-gov-west-1" - } + }, + "input_cost_per_token_batches": 3e-06, + "output_cost_per_token_batches": 1.5e-05, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-gov-west-1/index.json" }, "bedrock/us-gov-west-1/anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -68829,7 +68916,10 @@ "supports_sampling_params": false, "supports_tool_choice": true, "supports_vision": true, - "supports_xhigh_reasoning_effort": true + "supports_xhigh_reasoning_effort": true, + "input_cost_per_token_batches": 3e-06, + "output_cost_per_token_batches": 1.5e-05, + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-gov-east-1/index.json" }, "bedrock/us-gov-east-1/anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -69852,6 +69942,14 @@ "cache_read_input_token_cost": 3e-08, "source": "https://docs.perplexity.ai/docs/agent-api/models" }, + "perplexity/google/gemini-4-argon": { + "litellm_provider": "perplexity", + "mode": "responses", + "input_cost_per_token": 2e-06, + "output_cost_per_token": 1e-05, + "cache_read_input_token_cost": 1e-07, + "source": "https://docs.perplexity.ai/docs/agent-api/models" + }, "perplexity/google/gemini-3.6-flash": { "litellm_provider": "perplexity", "mode": "responses", @@ -69979,6 +70077,14 @@ "cache_read_input_token_cost": 1.15e-09, "source": "https://docs.perplexity.ai/docs/agent-api/models" }, + "perplexity/perplexity/nemotron-3.5-super-120b-a12b": { + "litellm_provider": "perplexity", + "mode": "responses", + "input_cost_per_token": 2e-07, + "output_cost_per_token": 5e-07, + "cache_read_input_token_cost": 4e-08, + "source": "https://docs.perplexity.ai/docs/agent-api/models" + }, "perplexity/perplexity/nemotron-3-ultra-550b-a55b": { "litellm_provider": "perplexity", "mode": "responses", @@ -70104,12 +70210,12 @@ "input_cost_per_token": 1.5e-06, "litellm_provider": "vertex_ai", "max_output_tokens": 57920, - "max_input_tokens": 57920, + "max_input_tokens": 131072, "mode": "chat", "output_cost_per_reasoning_token": 9e-06, "output_cost_per_token": 9e-06, "output_cost_per_video_token": 1.75e-05, - "source": "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing", + "source": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/omni-1-1-flash", "supported_endpoints": [ "/v1beta/interactions" ], @@ -74505,12 +74611,14 @@ "input_cost_per_token": 5.5e-06, "cache_read_input_token_cost": 5.5e-07, "output_cost_per_token": 2.75e-05, - "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/index.json", + "source": "https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/ap-northeast-1/index.json", "token_count_model": { "model": "anthropic.claude-opus-5", "litellm_provider": "bedrock_mantle", "region": "ap-southeast-4" - } + }, + "input_cost_per_token_batches": 2.75e-06, + "output_cost_per_token_batches": 1.375e-05 }, "apac.anthropic.claude-opus-5-5": { "bedrock_converse_supports_strict_tools": false, @@ -75107,7 +75215,7 @@ "supports_function_calling": true, "input_cost_per_token": 2e-06, "output_cost_per_token": 1e-05, - "cache_read_input_token_cost": 2e-07, + "cache_read_input_token_cost": 1e-07, "source": "https://docs.perplexity.ai/docs/agent-api/models" }, "perplexity/anthropic/claude-fable-5-1": { @@ -75155,6 +75263,9 @@ "output_cost_per_token": 5e-05, "cache_read_input_token_cost": 1e-06, "cache_creation_input_token_cost": 1.25e-05, + "input_cost_per_token_above_272k_tokens": 2e-05, + "output_cost_per_token_above_272k_tokens": 7.5e-05, + "cache_read_input_token_cost_above_272k_tokens": 2e-06, "source": "https://docs.perplexity.ai/docs/agent-api/models" }, "perplexity/openai/gpt-6-luna": { @@ -76622,6 +76733,9 @@ "input_cost_per_token": 1e-07, "output_cost_per_token": 5e-07, "cache_read_input_token_cost": 1e-08, + "input_cost_per_token_above_100k_tokens": 5e-07, + "output_cost_per_token_above_100k_tokens": 2.5e-06, + "cache_read_input_token_cost_above_100k_tokens": 5e-08, "source": "https://docs.perplexity.ai/docs/agent-api/models" }, "us-gov.anthropic.claude-haiku-5-5": { @@ -77096,6 +77210,7 @@ "web_search_billing_unit": "per_query" }, "vertex_ai/gemini-4-argon": { + "deprecation_date": "2027-10-13", "cache_read_input_token_cost": 1e-07, "cache_read_input_token_cost_batches": 5e-08, "cache_read_input_token_cost_flex": 5e-08, @@ -77504,9 +77619,10 @@ "max_output_tokens": 1048576, "mode": "chat", "output_cost_per_token": 1e-06, - "source": "https://tokenfactory.nebius.com/models/catalog/text2text/reflection%2Fbeam", + "source": "https://tokenfactory.nebius.com/api/public/models_info", "supports_function_calling": true, - "supports_reasoning": true + "supports_reasoning": true, + "cache_read_input_token_cost": 5e-08 }, "baseten/deepseek-ai/DeepSeek-V4.1-Flash-1017": { "cache_read_input_token_cost": 7e-09, @@ -81777,5 +81893,55 @@ "supports_response_schema": true, "supports_vision": true, "supports_xhigh_reasoning_effort": true + }, + "nebius/deepseek-ai/DeepSeek-V4.1-Flash-1017": { + "max_input_tokens": 1048576, + "input_cost_per_token": 3e-07, + "cache_read_input_token_cost": 6e-09, + "output_cost_per_token": 1.2e-06, + "litellm_provider": "nebius", + "mode": "chat", + "supports_vision": true, + "supports_reasoning": true, + "supports_function_calling": true, + "source": "https://tokenfactory.nebius.com/api/public/models_info" + }, + "nebius/meta-models/muse-spark-open": { + "max_input_tokens": 1048576, + "input_cost_per_token": 1.25e-06, + "cache_read_input_token_cost": 1.5e-07, + "output_cost_per_token": 4.25e-06, + "litellm_provider": "nebius", + "mode": "chat", + "supports_vision": true, + "supports_video_input": true, + "supports_reasoning": true, + "supports_function_calling": true, + "source": "https://tokenfactory.nebius.com/api/public/models_info" + }, + "nebius/MiniMaxAI/MiniMax-M3.1-Flash": { + "max_input_tokens": 1048576, + "input_cost_per_token": 1.5e-07, + "cache_read_input_token_cost": 1.5e-08, + "output_cost_per_token": 6e-07, + "litellm_provider": "nebius", + "mode": "chat", + "supports_vision": true, + "supports_reasoning": true, + "supports_function_calling": true, + "source": "https://tokenfactory.nebius.com/api/public/models_info" + }, + "nebius/nvidia/Nemotron-3.5-Super-VL": { + "max_input_tokens": 1048576, + "input_cost_per_token": 3e-07, + "cache_read_input_token_cost": 6e-08, + "output_cost_per_token": 9e-07, + "litellm_provider": "nebius", + "mode": "chat", + "supports_vision": true, + "supports_video_input": true, + "supports_reasoning": true, + "supports_function_calling": true, + "source": "https://tokenfactory.nebius.com/api/public/models_info" } } diff --git a/tests/unit/llms/azure_ai/test_azure_ai_foundry_catalog_model_metadata.py b/tests/unit/llms/azure_ai/test_azure_ai_foundry_catalog_model_metadata.py index 9b20192c3f2..a5b66efe7cd 100644 --- a/tests/unit/llms/azure_ai/test_azure_ai_foundry_catalog_model_metadata.py +++ b/tests/unit/llms/azure_ai/test_azure_ai_foundry_catalog_model_metadata.py @@ -22,6 +22,7 @@ TOKEN_PRICED_NAMES: Final = ( "cohere-command-a", "grok-4-20-reasoning", "grok-4-20-non-reasoning", + "claude-mythos-5-1", ) GROK_4_20_NAMES: Final = ("grok-4-20-reasoning", "grok-4-20-non-reasoning") CATALOG_NAMES: Final = TOKEN_PRICED_NAMES + ("whisper",)