diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 168cbeeade0..ff7b6b36dc8 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -15389,16 +15389,6 @@ "litellm_provider": "ollama", "mode": "completion" }, - "deepinfra/Austism/chronos-hermes-13b-v2": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 1.3e-07, - "output_cost_per_token": 1.3e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, "deepinfra/Gryphe/MythoMax-L2-13b": { "max_tokens": 4096, "max_input_tokens": 4096, @@ -15409,26 +15399,6 @@ "mode": "chat", "supports_tool_choice": true }, - "deepinfra/Gryphe/MythoMax-L2-13b-turbo": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 1.3e-07, - "output_cost_per_token": 1.3e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/KoboldAI/LLaMA2-13B-Tiefighter": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 1e-07, - "output_cost_per_token": 1e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, "deepinfra/NousResearch/Hermes-3-Llama-3.1-405B": { "max_tokens": 131072, "max_input_tokens": 131072, @@ -15447,78 +15417,18 @@ "output_cost_per_token": 2.8e-07, "litellm_provider": "deepinfra", "mode": "chat", - "supports_tool_choice": true - }, - "deepinfra/NovaSky-AI/Sky-T1-32B-Preview": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 1.2e-07, - "output_cost_per_token": 1.8e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/Phind/Phind-CodeLlama-34B-v2": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 6e-07, - "output_cost_per_token": 6e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, - "deepinfra/Qwen/QVQ-72B-Preview": { - "max_tokens": 32000, - "max_input_tokens": 32000, - "max_output_tokens": 32000, - "input_cost_per_token": 2.5e-07, - "output_cost_per_token": 5e-07, - "litellm_provider": "deepinfra", - "mode": "chat", "supports_tool_choice": false }, "deepinfra/Qwen/QwQ-32B": { "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 7.5e-08, - "output_cost_per_token": 1.5e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, - "deepinfra/Qwen/QwQ-32B-Preview": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 1.2e-07, - "output_cost_per_token": 1.8e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/Qwen/Qwen2-72B-Instruct": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 3.5e-07, + "input_cost_per_token": 1.5e-07, "output_cost_per_token": 4e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true }, - "deepinfra/Qwen/Qwen2-7B-Instruct": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 5.5e-08, - "output_cost_per_token": 5.5e-08, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, "deepinfra/Qwen/Qwen2.5-72B-Instruct": { "max_tokens": 32768, "max_input_tokens": 32768, @@ -15539,26 +15449,6 @@ "mode": "chat", "supports_tool_choice": false }, - "deepinfra/Qwen/Qwen2.5-Coder-32B-Instruct": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 6e-08, - "output_cost_per_token": 1.5e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/Qwen/Qwen2.5-Coder-7B": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 2.5e-08, - "output_cost_per_token": 5e-08, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, "deepinfra/Qwen/Qwen2.5-VL-32B-Instruct": { "max_tokens": 128000, "max_input_tokens": 128000, @@ -15645,30 +15535,11 @@ "max_output_tokens": 262144, "input_cost_per_token": 3e-07, "output_cost_per_token": 1.2e-06, + "cache_read_input_token_cost": 2.4e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true }, - "deepinfra/Sao10K/L3-70B-Euryale-v2.1": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 7e-07, - "output_cost_per_token": 8e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/Sao10K/L3-8B-Lunaris-v1": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 3e-08, - "output_cost_per_token": 6e-08, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, "deepinfra/Sao10K/L3-8B-Lunaris-v1-Turbo": { "max_tokens": 8192, "max_input_tokens": 8192, @@ -15715,6 +15586,7 @@ "max_output_tokens": 200000, "input_cost_per_token": 3.3e-06, "output_cost_per_token": 1.65e-05, + "cache_read_input_token_cost": 3.3e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -15739,67 +15611,15 @@ "mode": "chat", "supports_tool_choice": true }, - "deepinfra/bigcode/starcoder2-15b-instruct-v0.1": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 1.5e-07, - "output_cost_per_token": 1.5e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/cognitivecomputations/dolphin-2.6-mixtral-8x7b": { - "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": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, - "deepinfra/cognitivecomputations/dolphin-2.9.1-llama-3-70b": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 3.5e-07, - "output_cost_per_token": 4e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/deepinfra/airoboros-70b": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 7e-07, - "output_cost_per_token": 9e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, - "deepinfra/deepseek-ai/DeepSeek-Prover-V2-671B": { - "max_tokens": 163840, - "max_input_tokens": 163840, - "max_output_tokens": 163840, - "input_cost_per_token": 5e-07, - "output_cost_per_token": 2.18e-06, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false, - "supports_reasoning": true - }, "deepinfra/deepseek-ai/DeepSeek-R1": { "max_tokens": 163840, "max_input_tokens": 163840, "max_output_tokens": 163840, - "input_cost_per_token": 4.5e-07, - "output_cost_per_token": 2.15e-06, + "input_cost_per_token": 7e-07, + "output_cost_per_token": 2.4e-06, "litellm_provider": "deepinfra", "mode": "chat", - "supports_tool_choice": true, - "supports_reasoning": true + "supports_tool_choice": true }, "deepinfra/deepseek-ai/DeepSeek-R1-0528": { "max_tokens": 163840, @@ -15807,10 +15627,10 @@ "max_output_tokens": 163840, "input_cost_per_token": 5e-07, "output_cost_per_token": 2.15e-06, + "cache_read_input_token_cost": 4e-07, "litellm_provider": "deepinfra", "mode": "chat", - "supports_tool_choice": true, - "supports_reasoning": true + "supports_tool_choice": true }, "deepinfra/deepseek-ai/DeepSeek-R1-0528-Turbo": { "max_tokens": 32768, @@ -15820,8 +15640,7 @@ "output_cost_per_token": 3e-06, "litellm_provider": "deepinfra", "mode": "chat", - "supports_tool_choice": true, - "supports_reasoning": true + "supports_tool_choice": true }, "deepinfra/deepseek-ai/DeepSeek-R1-Distill-Llama-70B": { "max_tokens": 131072, @@ -15831,8 +15650,7 @@ "output_cost_per_token": 4e-07, "litellm_provider": "deepinfra", "mode": "chat", - "supports_tool_choice": false, - "supports_reasoning": true + "supports_tool_choice": false }, "deepinfra/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B": { "max_tokens": 131072, @@ -15842,19 +15660,17 @@ "output_cost_per_token": 1.5e-07, "litellm_provider": "deepinfra", "mode": "chat", - "supports_tool_choice": true, - "supports_reasoning": true + "supports_tool_choice": true }, "deepinfra/deepseek-ai/DeepSeek-R1-Turbo": { - "max_tokens": 163840, - "max_input_tokens": 163840, - "max_output_tokens": 163840, + "max_tokens": 40960, + "max_input_tokens": 40960, + "max_output_tokens": 40960, "input_cost_per_token": 1e-06, "output_cost_per_token": 3e-06, "litellm_provider": "deepinfra", "mode": "chat", - "supports_tool_choice": true, - "supports_reasoning": true + "supports_tool_choice": true }, "deepinfra/deepseek-ai/DeepSeek-V3": { "max_tokens": 163840, @@ -15864,8 +15680,7 @@ "output_cost_per_token": 8.9e-07, "litellm_provider": "deepinfra", "mode": "chat", - "supports_tool_choice": true, - "supports_reasoning": true + "supports_tool_choice": true }, "deepinfra/deepseek-ai/DeepSeek-V3-0324": { "max_tokens": 163840, @@ -15873,59 +15688,18 @@ "max_output_tokens": 163840, "input_cost_per_token": 2.8e-07, "output_cost_per_token": 8.8e-07, + "cache_read_input_token_cost": 2.24e-07, "litellm_provider": "deepinfra", "mode": "chat", - "supports_tool_choice": true, - "supports_reasoning": true - }, - "deepinfra/deepseek-ai/DeepSeek-V3-0324-Turbo": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 1e-06, - "output_cost_per_token": 3e-06, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true, - "supports_reasoning": true + "supports_tool_choice": true }, "deepinfra/deepseek-ai/DeepSeek-V3.1": { "max_tokens": 163840, "max_input_tokens": 163840, "max_output_tokens": 163840, - "input_cost_per_token": 3e-07, + "input_cost_per_token": 2.7e-07, "output_cost_per_token": 1e-06, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false, - "supports_reasoning": true - }, - "deepinfra/google/codegemma-7b-it": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 7e-08, - "output_cost_per_token": 7e-08, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/google/gemini-1.5-flash": { - "max_tokens": 1000000, - "max_input_tokens": 1000000, - "max_output_tokens": 1000000, - "input_cost_per_token": 7.5e-08, - "output_cost_per_token": 3e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, - "deepinfra/google/gemini-1.5-flash-8b": { - "max_tokens": 1000000, - "max_input_tokens": 1000000, - "max_output_tokens": 1000000, - "input_cost_per_token": 3.75e-08, - "output_cost_per_token": 1.5e-07, + "cache_read_input_token_cost": 2.16e-07, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -15960,36 +15734,6 @@ "mode": "chat", "supports_tool_choice": true }, - "deepinfra/google/gemma-1.1-7b-it": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 7e-08, - "output_cost_per_token": 7e-08, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, - "deepinfra/google/gemma-2-27b-it": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 2.7e-07, - "output_cost_per_token": 2.7e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/google/gemma-2-9b-it": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 3e-08, - "output_cost_per_token": 6e-08, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, "deepinfra/google/gemma-3-12b-it": { "max_tokens": 131072, "max_input_tokens": 131072, @@ -16014,48 +15758,8 @@ "max_tokens": 131072, "max_input_tokens": 131072, "max_output_tokens": 131072, - "input_cost_per_token": 2e-08, - "output_cost_per_token": 4e-08, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, - "deepinfra/lizpreciatior/lzlv_70b_fp16_hf": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 3.5e-07, - "output_cost_per_token": 4e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/mattshumer/Reflection-Llama-3.1-70B": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 3.5e-07, - "output_cost_per_token": 4e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/meta-llama/Llama-2-13b-chat-hf": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 1.3e-07, - "output_cost_per_token": 1.3e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, - "deepinfra/meta-llama/Llama-2-70b-chat-hf": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 6.4e-07, - "output_cost_per_token": 8e-07, + "input_cost_per_token": 4e-08, + "output_cost_per_token": 8e-08, "litellm_provider": "deepinfra", "mode": "chat", "supports_tool_choice": true @@ -16070,16 +15774,6 @@ "mode": "chat", "supports_tool_choice": false }, - "deepinfra/meta-llama/Llama-3.2-1B-Instruct": { - "max_tokens": 131072, - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 5e-09, - "output_cost_per_token": 1e-08, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, "deepinfra/meta-llama/Llama-3.2-3B-Instruct": { "max_tokens": 131072, "max_input_tokens": 131072, @@ -16090,16 +15784,6 @@ "mode": "chat", "supports_tool_choice": true }, - "deepinfra/meta-llama/Llama-3.2-90B-Vision-Instruct": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 3.5e-07, - "output_cost_per_token": 4e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, "deepinfra/meta-llama/Llama-3.3-70B-Instruct": { "max_tokens": 131072, "max_input_tokens": 131072, @@ -16130,16 +15814,6 @@ "mode": "chat", "supports_tool_choice": true }, - "deepinfra/meta-llama/Llama-4-Maverick-17B-128E-Instruct-Turbo": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 5e-07, - "output_cost_per_token": 5e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, "deepinfra/meta-llama/Llama-4-Scout-17B-16E-Instruct": { "max_tokens": 327680, "max_input_tokens": 327680, @@ -16170,16 +15844,6 @@ "mode": "chat", "supports_tool_choice": false }, - "deepinfra/meta-llama/Meta-Llama-3-70B-Instruct": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 3e-07, - "output_cost_per_token": 4e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, "deepinfra/meta-llama/Meta-Llama-3-8B-Instruct": { "max_tokens": 8192, "max_input_tokens": 8192, @@ -16190,16 +15854,6 @@ "mode": "chat", "supports_tool_choice": true }, - "deepinfra/meta-llama/Meta-Llama-3.1-405B-Instruct": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 8e-07, - "output_cost_per_token": 8e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, "deepinfra/meta-llama/Meta-Llama-3.1-70B-Instruct": { "max_tokens": 131072, "max_input_tokens": 131072, @@ -16240,36 +15894,6 @@ "mode": "chat", "supports_tool_choice": true }, - "deepinfra/microsoft/Phi-3-medium-4k-instruct": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 1.4e-07, - "output_cost_per_token": 1.4e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/microsoft/Phi-4-multimodal-instruct": { - "max_tokens": 131072, - "max_input_tokens": 131072, - "max_output_tokens": 131072, - "input_cost_per_token": 5e-08, - "output_cost_per_token": 1e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/microsoft/WizardLM-2-7B": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 5.5e-08, - "output_cost_per_token": 5.5e-08, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, "deepinfra/microsoft/WizardLM-2-8x22B": { "max_tokens": 65536, "max_input_tokens": 65536, @@ -16290,66 +15914,6 @@ "mode": "chat", "supports_tool_choice": true }, - "deepinfra/microsoft/phi-4-reasoning-plus": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 7e-08, - "output_cost_per_token": 3.5e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/mistralai/Devstral-Small-2505": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 6e-08, - "output_cost_per_token": 1.2e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, - "deepinfra/mistralai/Devstral-Small-2507": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 7e-08, - "output_cost_per_token": 2.8e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, - "deepinfra/mistralai/Mistral-7B-Instruct-v0.1": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 5.5e-08, - "output_cost_per_token": 5.5e-08, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, - "deepinfra/mistralai/Mistral-7B-Instruct-v0.2": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 5.5e-08, - "output_cost_per_token": 5.5e-08, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/mistralai/Mistral-7B-Instruct-v0.3": { - "max_tokens": 32768, - "max_input_tokens": 32768, - "max_output_tokens": 32768, - "input_cost_per_token": 2.8e-08, - "output_cost_per_token": 5.4e-08, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, "deepinfra/mistralai/Mistral-Nemo-Instruct-2407": { "max_tokens": 131072, "max_input_tokens": 131072, @@ -16370,16 +15934,6 @@ "mode": "chat", "supports_tool_choice": true }, - "deepinfra/mistralai/Mistral-Small-3.1-24B-Instruct-2503": { - "max_tokens": 128000, - "max_input_tokens": 128000, - "max_output_tokens": 128000, - "input_cost_per_token": 5e-08, - "output_cost_per_token": 1e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, "deepinfra/mistralai/Mistral-Small-3.2-24B-Instruct-2506": { "max_tokens": 128000, "max_input_tokens": 128000, @@ -16390,16 +15944,6 @@ "mode": "chat", "supports_tool_choice": true }, - "deepinfra/mistralai/Mixtral-8x22B-Instruct-v0.1": { - "max_tokens": 65536, - "max_input_tokens": 65536, - "max_output_tokens": 65536, - "input_cost_per_token": 6.5e-07, - "output_cost_per_token": 6.5e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, "deepinfra/mistralai/Mixtral-8x7B-Instruct-v0.1": { "max_tokens": 32768, "max_input_tokens": 32768, @@ -16430,16 +15974,6 @@ "mode": "chat", "supports_tool_choice": true }, - "deepinfra/nvidia/Nemotron-4-340B-Instruct": { - "max_tokens": 4096, - "max_input_tokens": 4096, - "max_output_tokens": 4096, - "input_cost_per_token": 4.2e-06, - "output_cost_per_token": 4.2e-06, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, "deepinfra/openai/gpt-oss-120b": { "max_tokens": 131072, "max_input_tokens": 131072, @@ -16460,36 +15994,6 @@ "mode": "chat", "supports_tool_choice": true }, - "deepinfra/openbmb/MiniCPM-Llama3-V-2_5": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 3.4e-07, - "output_cost_per_token": 3.4e-07, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/openchat/openchat-3.6-8b": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 5.5e-08, - "output_cost_per_token": 5.5e-08, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": false - }, - "deepinfra/openchat/openchat_3.5": { - "max_tokens": 8192, - "max_input_tokens": 8192, - "max_output_tokens": 8192, - "input_cost_per_token": 5.5e-08, - "output_cost_per_token": 5.5e-08, - "litellm_provider": "deepinfra", - "mode": "chat", - "supports_tool_choice": true - }, "deepinfra/zai-org/GLM-4.5": { "max_tokens": 131072, "max_input_tokens": 131072, @@ -20998,4 +20502,4 @@ "notes": "Volcengine Doubao embedding model - text-240715 version with 2560 dimensions" } } -} \ No newline at end of file +} diff --git a/litellm/types/llms/databricks.py b/litellm/types/llms/databricks.py index c362d065694..c484e80ada3 100644 --- a/litellm/types/llms/databricks.py +++ b/litellm/types/llms/databricks.py @@ -51,7 +51,7 @@ AllDatabricksContentValues = Union[str, List[AllDatabricksContentListValues]] class DatabricksFunction(TypedDict, total=False): name: Required[str] - description: dict | str + description: Union[dict, str] parameters: dict strict: bool