diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index d42762355c..36d1a59aec 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -3423,25 +3423,24 @@ "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": 7.900000000000001e-07, "litellm_provider": "groq", "mode": "chat", "supports_function_calling": true, @@ -3452,11 +3451,21 @@ "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": 2.0000000000000002e-07, + "output_cost_per_token": 2.0000000000000002e-07, + "litellm_provider": "groq", + "mode": "chat" }, "groq/llama2-70b-4096": { "max_tokens": 4096, @@ -3474,106 +3483,106 @@ "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": 5.0000000000000004e-08, + "output_cost_per_token": 8e-08, "litellm_provider": "groq", - "mode": "chat", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true + "mode": "chat" }, "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": 7.900000000000001e-07, "litellm_provider": "groq", - "mode": "chat", - "supports_function_calling": true, - "supports_response_schema": true, - "supports_tool_choice": true + "mode": "chat" }, "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": 5.0000000000000004e-08, + "output_cost_per_token": 8e-08, "litellm_provider": "groq", "mode": "chat", "supports_function_calling": true, @@ -3584,13 +3593,14 @@ "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, @@ -3604,83 +3614,141 @@ "supports_response_schema": true, "supports_tool_choice": true }, - "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, + "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": 3.4000000000000003e-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": 2.0000000000000002e-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": 7.900000000000001e-07, + "output_cost_per_token": 7.900000000000001e-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": 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, + "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": 2.0000000000000002e-07, + "output_cost_per_token": 2.0000000000000002e-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": "2025-1-6" }, "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": "2025-1-6" + }, + "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": 3.0833333333333335e-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": 1.1111111111111112e-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": 5.555555555555556e-06, + "output_cost_per_second": 0.0, + "litellm_provider": "groq", + "mode": "audio_transcription" }, "cerebras/llama3.1-8b": { "max_tokens": 128000,