scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-03 02:22:24 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 81

fix: resolve floating point artifacts and add trailing newline

Browse source
This commit is contained in:
Kutsukey 2026-05-01 11:28:04 +03:00
parent 9a3f7a616a
commit b4396ec771
2 changed files with 174 additions and 234 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

168
litellm/model_prices_and_context_window_backup.json
View file

@ -9477,7 +9477,6 @@
"us": 1.1,
"fast": 6.0
},
"supports_max_reasoning_effort": true,
"supports_minimal_reasoning_effort": true
},
"claude-opus-4-7-20260416": {
@ -9512,7 +9511,6 @@
"us": 1.1,
"fast": 6.0
},
"supports_max_reasoning_effort": true,
"supports_minimal_reasoning_effort": true
},
"claude-sonnet-4-20250514": {
@ -10698,17 +10696,17 @@
"source": "https://www.databricks.com/product/pricing/foundation-model-serving"
},
"databricks/databricks-claude-3-7-sonnet": {
"input_cost_per_token": 2.9999900000000002e-06,
"input_cost_per_token": 2.99999e-06,
"input_dbu_cost_per_token": 4.2857e-05,
"litellm_provider": "databricks",
"max_input_tokens": 200000,
"max_output_tokens": 128000,
"max_tokens": 128000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 1.5000020000000002e-05,
"output_cost_per_token": 1.500002e-05,
"output_dbu_cost_per_token": 0.000214286,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving",
"supports_assistant_prefill": true,
@ -10724,7 +10722,7 @@
"max_output_tokens": 64000,
"max_tokens": 64000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 5.00003e-06,
@ -10736,17 +10734,17 @@
"supports_tool_choice": true
},
"databricks/databricks-claude-opus-4": {
"input_cost_per_token": 1.5000020000000002e-05,
"input_cost_per_token": 1.500002e-05,
"input_dbu_cost_per_token": 0.000214286,
"litellm_provider": "databricks",
"max_input_tokens": 200000,
"max_output_tokens": 32000,
"max_tokens": 32000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 7.500003000000001e-05,
"output_cost_per_token": 7.500003e-05,
"output_dbu_cost_per_token": 0.001071429,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving",
"supports_assistant_prefill": true,
@ -10755,17 +10753,17 @@
"supports_tool_choice": true
},
"databricks/databricks-claude-opus-4-1": {
"input_cost_per_token": 1.5000020000000002e-05,
"input_cost_per_token": 1.500002e-05,
"input_dbu_cost_per_token": 0.000214286,
"litellm_provider": "databricks",
"max_input_tokens": 200000,
"max_output_tokens": 32000,
"max_tokens": 32000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 7.500003000000001e-05,
"output_cost_per_token": 7.500003e-05,
"output_dbu_cost_per_token": 0.001071429,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving",
"supports_assistant_prefill": true,
@ -10781,10 +10779,10 @@
"max_output_tokens": 64000,
"max_tokens": 64000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 2.5000010000000002e-05,
"output_cost_per_token": 2.500001e-05,
"output_dbu_cost_per_token": 0.000357143,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving",
"supports_assistant_prefill": true,
@ -10793,17 +10791,17 @@
"supports_tool_choice": true
},
"databricks/databricks-claude-sonnet-4": {
"input_cost_per_token": 2.9999900000000002e-06,
"input_cost_per_token": 2.99999e-06,
"input_dbu_cost_per_token": 4.2857e-05,
"litellm_provider": "databricks",
"max_input_tokens": 200000,
"max_output_tokens": 64000,
"max_tokens": 64000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 1.5000020000000002e-05,
"output_cost_per_token": 1.500002e-05,
"output_dbu_cost_per_token": 0.000214286,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving",
"supports_assistant_prefill": true,
@ -10812,17 +10810,17 @@
"supports_tool_choice": true
},
"databricks/databricks-claude-sonnet-4-1": {
"input_cost_per_token": 2.9999900000000002e-06,
"input_cost_per_token": 2.99999e-06,
"input_dbu_cost_per_token": 4.2857e-05,
"litellm_provider": "databricks",
"max_input_tokens": 200000,
"max_output_tokens": 64000,
"max_tokens": 64000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 1.5000020000000002e-05,
"output_cost_per_token": 1.500002e-05,
"output_dbu_cost_per_token": 0.000214286,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving",
"supports_assistant_prefill": true,
@ -10831,17 +10829,17 @@
"supports_tool_choice": true
},
"databricks/databricks-claude-sonnet-4-5": {
"input_cost_per_token": 2.9999900000000002e-06,
"input_cost_per_token": 2.99999e-06,
"input_dbu_cost_per_token": 4.2857e-05,
"litellm_provider": "databricks",
"max_input_tokens": 200000,
"max_output_tokens": 64000,
"max_tokens": 64000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 1.5000020000000002e-05,
"output_cost_per_token": 1.500002e-05,
"output_dbu_cost_per_token": 0.000214286,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving",
"supports_assistant_prefill": true,
@ -10850,14 +10848,14 @@
"supports_tool_choice": true
},
"databricks/databricks-gemini-2-5-flash": {
"input_cost_per_token": 3.0001999999999996e-07,
"input_dbu_cost_per_token": 4.285999999999999e-06,
"input_cost_per_token": 3.0002e-07,
"input_dbu_cost_per_token": 4.286e-06,
"litellm_provider": "databricks",
"max_input_tokens": 1048576,
"max_output_tokens": 65535,
"max_tokens": 65535,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 2.49998e-06,
@ -10874,24 +10872,24 @@
"max_output_tokens": 65536,
"max_tokens": 65536,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 9.999990000000002e-06,
"output_cost_per_token": 9.99999e-06,
"output_dbu_cost_per_token": 0.000142857,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving",
"supports_function_calling": true,
"supports_tool_choice": true
},
"databricks/databricks-gemma-3-12b": {
"input_cost_per_token": 1.5000999999999998e-07,
"input_dbu_cost_per_token": 2.1429999999999996e-06,
"input_cost_per_token": 1.5001e-07,
"input_dbu_cost_per_token": 2.143e-06,
"litellm_provider": "databricks",
"max_input_tokens": 128000,
"max_output_tokens": 32000,
"max_tokens": 32000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 5.0001e-07,
@ -10906,10 +10904,10 @@
"max_output_tokens": 128000,
"max_tokens": 128000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 9.999990000000002e-06,
"output_cost_per_token": 9.99999e-06,
"output_dbu_cost_per_token": 0.000142857,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving"
},
@ -10921,25 +10919,25 @@
"max_output_tokens": 128000,
"max_tokens": 128000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 9.999990000000002e-06,
"output_cost_per_token": 9.99999e-06,
"output_dbu_cost_per_token": 0.000142857,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving"
},
"databricks/databricks-gpt-5-mini": {
"input_cost_per_token": 2.4997000000000006e-07,
"input_cost_per_token": 2.4997e-07,
"input_dbu_cost_per_token": 3.571e-06,
"litellm_provider": "databricks",
"max_input_tokens": 272000,
"max_output_tokens": 128000,
"max_tokens": 128000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 1.9999700000000004e-06,
"output_cost_per_token": 1.99997e-06,
"output_dbu_cost_per_token": 2.8571e-05,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving"
},
@ -10951,22 +10949,22 @@
"max_output_tokens": 128000,
"max_tokens": 128000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 3.9998000000000007e-07,
"output_dbu_cost_per_token": 5.714000000000001e-06,
"output_cost_per_token": 3.9998e-07,
"output_dbu_cost_per_token": 5.714e-06,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving"
},
"databricks/databricks-gpt-oss-120b": {
"input_cost_per_token": 1.5000999999999998e-07,
"input_dbu_cost_per_token": 2.1429999999999996e-06,
"input_cost_per_token": 1.5001e-07,
"input_dbu_cost_per_token": 2.143e-06,
"litellm_provider": "databricks",
"max_input_tokens": 131072,
"max_output_tokens": 131072,
"max_tokens": 131072,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 5.9997e-07,
@ -10981,15 +10979,15 @@
"max_output_tokens": 131072,
"max_tokens": 131072,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 3.0001999999999996e-07,
"output_dbu_cost_per_token": 4.285999999999999e-06,
"output_cost_per_token": 3.0002e-07,
"output_dbu_cost_per_token": 4.286e-06,
"source": "https://www.databricks.com/product/pricing/foundation-model-serving"
},
"databricks/databricks-gte-large-en": {
"input_cost_per_token": 1.2999000000000001e-07,
"input_cost_per_token": 1.2999e-07,
"input_dbu_cost_per_token": 1.857e-06,
"litellm_provider": "databricks",
"max_input_tokens": 8192,
@ -11014,7 +11012,7 @@
"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 1.5000300000000002e-06,
"output_cost_per_token": 1.50003e-06,
"output_dbu_cost_per_token": 2.1429e-05,
"source": "https://www.databricks.com/product/pricing/foundation-model-serving",
"supports_tool_choice": true
@ -11030,7 +11028,7 @@
"notes": "Databricks documentation now provides both DBU costs (_dbu_cost_per_token) and dollar costs(_cost_per_token)."
},
"mode": "chat",
"output_cost_per_token": 1.5000300000000002e-06,
"output_cost_per_token": 1.50003e-06,
"output_dbu_cost_per_token": 2.1429e-05,
"source": "https://www.databricks.com/product/pricing/foundation-model-serving",
"supports_tool_choice": true
@ -11046,24 +11044,24 @@
"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 1.5000020000000002e-05,
"output_cost_per_token": 1.500002e-05,
"output_dbu_cost_per_token": 0.000214286,
"source": "https://www.databricks.com/product/pricing/foundation-model-serving",
"supports_tool_choice": true
},
"databricks/databricks-meta-llama-3-1-8b-instruct": {
"input_cost_per_token": 1.5000999999999998e-07,
"input_dbu_cost_per_token": 2.1429999999999996e-06,
"input_cost_per_token": 1.5001e-07,
"input_dbu_cost_per_token": 2.143e-06,
"litellm_provider": "databricks",
"max_input_tokens": 200000,
"max_output_tokens": 128000,
"max_tokens": 128000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 4.5003000000000007e-07,
"output_dbu_cost_per_token": 6.429000000000001e-06,
"output_cost_per_token": 4.5003e-07,
"output_dbu_cost_per_token": 6.429e-06,
"source": "https://www.databricks.com/product/pricing/foundation-model-serving"
},
"databricks/databricks-meta-llama-3-3-70b-instruct": {
@ -11077,7 +11075,7 @@
"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 1.5000300000000002e-06,
"output_cost_per_token": 1.50003e-06,
"output_dbu_cost_per_token": 2.1429e-05,
"source": "https://www.databricks.com/product/pricing/foundation-model-serving",
"supports_tool_choice": true
@ -11093,7 +11091,7 @@
"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 2.9999900000000002e-06,
"output_cost_per_token": 2.99999e-06,
"output_dbu_cost_per_token": 4.2857e-05,
"source": "https://www.databricks.com/product/pricing/foundation-model-serving",
"supports_tool_choice": true
@ -16827,7 +16825,7 @@
"max_video_length": 1,
"max_videos_per_prompt": 10,
"metadata": {
"notes": "Rate limits not documented for gemini-exp-1114. Assuming same as gemini-1.5-pro.",
"notes": "Rate limits not documented for gemini-exp-1114 Assuming same as gemini-1.5-pro.",
"supports_tool_choice": true
},
"mode": "chat",
@ -16856,7 +16854,7 @@
"max_video_length": 1,
"max_videos_per_prompt": 10,
"metadata": {
"notes": "Rate limits not documented for gemini-exp-1206. Assuming same as gemini-1.5-pro.",
"notes": "Rate limits not documented for gemini-exp-1206 Assuming same as gemini-1.5-pro.",
"supports_tool_choice": true
},
"mode": "chat",
@ -39622,7 +39620,7 @@
"max_input_tokens": 198000,
"max_output_tokens": 32000,
"input_cost_per_token": 1e-06,
"output_cost_per_token": 3.2000000000000003e-06,
"output_cost_per_token": 3.2e-06,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -39667,7 +39665,7 @@
"max_input_tokens": 200000,
"max_output_tokens": 24000,
"input_cost_per_token": 1.4e-07,
"output_cost_per_token": 8.000000000000001e-07,
"output_cost_per_token": 8e-07,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -39726,8 +39724,8 @@
"max_tokens": 8192,
"max_input_tokens": 32000,
"max_output_tokens": 8192,
"input_cost_per_token": 2.0000000000000002e-07,
"output_cost_per_token": 9.000000000000001e-07,
"input_cost_per_token": 2e-07,
"output_cost_per_token": 9e-07,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -39741,8 +39739,8 @@
"max_tokens": 8192,
"max_input_tokens": 128000,
"max_output_tokens": 8192,
"input_cost_per_token": 2.0000000000000002e-07,
"output_cost_per_token": 9.000000000000001e-07,
"input_cost_per_token": 2e-07,
"output_cost_per_token": 9e-07,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -39801,7 +39799,7 @@
"max_tokens": 32768,
"max_input_tokens": 256000,
"max_output_tokens": 32768,
"input_cost_per_token": 1.0000000000000001e-07,
"input_cost_per_token": 1e-07,
"output_cost_per_token": 1.5e-07,
"litellm_provider": "venice",
"mode": "chat",
@ -39846,7 +39844,7 @@
"max_tokens": 16384,
"max_input_tokens": 128000,
"max_output_tokens": 16384,
"input_cost_per_token": 4.5000000000000003e-07,
"input_cost_per_token": 4.5e-07,
"output_cost_per_token": 3.5e-06,
"litellm_provider": "venice",
"mode": "chat",
@ -39877,7 +39875,7 @@
"max_input_tokens": 256000,
"max_output_tokens": 16384,
"input_cost_per_token": 3.5e-07,
"output_cost_per_token": 1.8999999999999998e-06,
"output_cost_per_token": 1.9e-06,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -39982,7 +39980,7 @@
"max_input_tokens": 198000,
"max_output_tokens": 16384,
"input_cost_per_token": 1.2e-07,
"output_cost_per_token": 2.0000000000000002e-07,
"output_cost_per_token": 2e-07,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -40011,8 +40009,8 @@
"max_tokens": 30000,
"max_input_tokens": 1000000,
"max_output_tokens": 30000,
"input_cost_per_token": 2.3000000000000002e-07,
"output_cost_per_token": 5.699999999999999e-07,
"input_cost_per_token": 2.3e-07,
"output_cost_per_token": 5.7e-07,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -40206,7 +40204,7 @@
"max_tokens": 64000,
"max_input_tokens": 1000000,
"max_output_tokens": 64000,
"input_cost_per_token": 3.6000000000000003e-06,
"input_cost_per_token": 3.6e-06,
"output_cost_per_token": 1.8e-05,
"litellm_provider": "venice",
"mode": "chat",
@ -40267,7 +40265,7 @@
"max_input_tokens": 256000,
"max_output_tokens": 65536,
"input_cost_per_token": 7.5e-07,
"output_cost_per_token": 3.2000000000000003e-06,
"output_cost_per_token": 3.2e-06,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -40297,7 +40295,7 @@
"max_input_tokens": 1000000,
"max_output_tokens": 32768,
"input_cost_per_token": 1.73e-06,
"output_cost_per_token": 3.7959999999999997e-06,
"output_cost_per_token": 3.796e-06,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -40311,7 +40309,7 @@
"max_tokens": 32768,
"max_input_tokens": 1000000,
"max_output_tokens": 32768,
"input_cost_per_token": 1.7000000000000001e-07,
"input_cost_per_token": 1.7e-07,
"output_cost_per_token": 3.5e-07,
"litellm_provider": "venice",
"mode": "chat",
@ -40386,7 +40384,7 @@
"max_tokens": 65536,
"max_input_tokens": 256000,
"max_output_tokens": 65536,
"input_cost_per_token": 2.1899999999999998e-06,
"input_cost_per_token": 2.19e-06,
"output_cost_per_token": 1.75e-05,
"litellm_provider": "venice",
"mode": "chat",
@ -40401,7 +40399,7 @@
"max_tokens": 65536,
"max_input_tokens": 256000,
"max_output_tokens": 65536,
"input_cost_per_token": 2.1899999999999998e-06,
"input_cost_per_token": 2.19e-06,
"output_cost_per_token": 1.75e-05,
"litellm_provider": "venice",
"mode": "chat",
@ -40416,7 +40414,7 @@
"max_tokens": 128000,
"max_input_tokens": 400000,
"max_output_tokens": 128000,
"input_cost_per_token": 2.1899999999999998e-06,
"input_cost_per_token": 2.19e-06,
"output_cost_per_token": 1.75e-05,
"litellm_provider": "venice",
"mode": "chat",
@ -40536,7 +40534,7 @@
"max_tokens": 32768,
"max_input_tokens": 198000,
"max_output_tokens": 32768,
"input_cost_per_token": 3.4000000000000003e-07,
"input_cost_per_token": 3.4e-07,
"output_cost_per_token": 1.19e-06,
"litellm_provider": "venice",
"mode": "chat",
@ -40597,7 +40595,7 @@
"max_input_tokens": 256000,
"max_output_tokens": 32768,
"input_cost_per_token": 1.4e-07,
"output_cost_per_token": 8.000000000000001e-07,
"output_cost_per_token": 8.0e-07,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -40671,7 +40669,7 @@
"max_tokens": 32768,
"max_input_tokens": 128000,
"max_output_tokens": 32768,
"input_cost_per_token": 5.0000000000000004e-08,
"input_cost_per_token": 5e-08,
"output_cost_per_token": 1.9e-07,
"litellm_provider": "venice",
"mode": "chat",
@ -40701,7 +40699,7 @@
"max_tokens": 4096,
"max_input_tokens": 32000,
"max_output_tokens": 4096,
"input_cost_per_token": 5.0000000000000004e-08,
"input_cost_per_token": 5.0e-08,
"output_cost_per_token": 1.3e-07,
"litellm_provider": "venice",
"mode": "chat",
@ -40732,7 +40730,7 @@
"max_input_tokens": 128000,
"max_output_tokens": 4096,
"input_cost_per_token": 2.5e-07,
"output_cost_per_token": 9.000000000000001e-07,
"output_cost_per_token": 9.0e-07,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -40787,4 +40785,4 @@
"supports_vision": true,
"supports_web_search": true
}
}
}

240
model_prices_and_context_window.json
View file

@ -8092,17 +8092,16 @@
},
"bedrock/us-east-1/minimax.minimax-m2.5": {
"input_cost_per_token": 3e-07,
"output_cost_per_token": 1.2e-06,
"litellm_provider": "bedrock",
"max_input_tokens": 1000000,
"max_output_tokens": 8192,
"max_tokens": 8192,
"mode": "chat",
"source": "https://aws.amazon.com/bedrock/pricing/",
"supports_function_calling": true,
"supports_reasoning": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"output_cost_per_token": 1.2e-06
"source": "https://aws.amazon.com/bedrock/pricing/"
},
"bedrock/us-east-1/moonshotai.kimi-k2-thinking": {
"input_cost_per_token": 6e-07,
@ -8710,17 +8709,16 @@
},
"bedrock/us-west-2/minimax.minimax-m2.5": {
"input_cost_per_token": 3e-07,
"output_cost_per_token": 1.2e-06,
"litellm_provider": "bedrock",
"max_input_tokens": 1000000,
"max_output_tokens": 8192,
"max_tokens": 8192,
"mode": "chat",
"source": "https://aws.amazon.com/bedrock/pricing/",
"supports_function_calling": true,
"supports_reasoning": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"output_cost_per_token": 1.2e-06
"source": "https://aws.amazon.com/bedrock/pricing/"
},
"bedrock/us-west-2/moonshotai.kimi-k2-thinking": {
"input_cost_per_token": 6e-07,
@ -9491,7 +9489,6 @@
"us": 1.1,
"fast": 6.0
},
"supports_max_reasoning_effort": true,
"supports_minimal_reasoning_effort": true
},
"claude-opus-4-7-20260416": {
@ -9526,7 +9523,6 @@
"us": 1.1,
"fast": 6.0
},
"supports_max_reasoning_effort": true,
"supports_minimal_reasoning_effort": true
},
"claude-sonnet-4-20250514": {
@ -10712,17 +10708,17 @@
"source": "https://www.databricks.com/product/pricing/foundation-model-serving"
},
"databricks/databricks-claude-3-7-sonnet": {
"input_cost_per_token": 2.9999900000000002e-06,
"input_cost_per_token": 2.99999e-06,
"input_dbu_cost_per_token": 4.2857e-05,
"litellm_provider": "databricks",
"max_input_tokens": 200000,
"max_output_tokens": 128000,
"max_tokens": 128000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 1.5000020000000002e-05,
"output_cost_per_token": 1.500002e-05,
"output_dbu_cost_per_token": 0.000214286,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving",
"supports_assistant_prefill": true,
@ -10738,7 +10734,7 @@
"max_output_tokens": 64000,
"max_tokens": 64000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 5.00003e-06,
@ -10750,17 +10746,17 @@
"supports_tool_choice": true
},
"databricks/databricks-claude-opus-4": {
"input_cost_per_token": 1.5000020000000002e-05,
"input_cost_per_token": 1.500002e-05,
"input_dbu_cost_per_token": 0.000214286,
"litellm_provider": "databricks",
"max_input_tokens": 200000,
"max_output_tokens": 32000,
"max_tokens": 32000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 7.500003000000001e-05,
"output_cost_per_token": 7.500003e-05,
"output_dbu_cost_per_token": 0.001071429,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving",
"supports_assistant_prefill": true,
@ -10769,17 +10765,17 @@
"supports_tool_choice": true
},
"databricks/databricks-claude-opus-4-1": {
"input_cost_per_token": 1.5000020000000002e-05,
"input_cost_per_token": 1.500002e-05,
"input_dbu_cost_per_token": 0.000214286,
"litellm_provider": "databricks",
"max_input_tokens": 200000,
"max_output_tokens": 32000,
"max_tokens": 32000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 7.500003000000001e-05,
"output_cost_per_token": 7.500003e-05,
"output_dbu_cost_per_token": 0.001071429,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving",
"supports_assistant_prefill": true,
@ -10795,10 +10791,10 @@
"max_output_tokens": 64000,
"max_tokens": 64000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 2.5000010000000002e-05,
"output_cost_per_token": 2.500001e-05,
"output_dbu_cost_per_token": 0.000357143,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving",
"supports_assistant_prefill": true,
@ -10807,17 +10803,17 @@
"supports_tool_choice": true
},
"databricks/databricks-claude-sonnet-4": {
"input_cost_per_token": 2.9999900000000002e-06,
"input_cost_per_token": 2.99999e-06,
"input_dbu_cost_per_token": 4.2857e-05,
"litellm_provider": "databricks",
"max_input_tokens": 200000,
"max_output_tokens": 64000,
"max_tokens": 64000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 1.5000020000000002e-05,
"output_cost_per_token": 1.500002e-05,
"output_dbu_cost_per_token": 0.000214286,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving",
"supports_assistant_prefill": true,
@ -10826,17 +10822,17 @@
"supports_tool_choice": true
},
"databricks/databricks-claude-sonnet-4-1": {
"input_cost_per_token": 2.9999900000000002e-06,
"input_cost_per_token": 2.99999e-06,
"input_dbu_cost_per_token": 4.2857e-05,
"litellm_provider": "databricks",
"max_input_tokens": 200000,
"max_output_tokens": 64000,
"max_tokens": 64000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 1.5000020000000002e-05,
"output_cost_per_token": 1.500002e-05,
"output_dbu_cost_per_token": 0.000214286,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving",
"supports_assistant_prefill": true,
@ -10845,17 +10841,17 @@
"supports_tool_choice": true
},
"databricks/databricks-claude-sonnet-4-5": {
"input_cost_per_token": 2.9999900000000002e-06,
"input_cost_per_token": 2.99999e-06,
"input_dbu_cost_per_token": 4.2857e-05,
"litellm_provider": "databricks",
"max_input_tokens": 200000,
"max_output_tokens": 64000,
"max_tokens": 64000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 1.5000020000000002e-05,
"output_cost_per_token": 1.500002e-05,
"output_dbu_cost_per_token": 0.000214286,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving",
"supports_assistant_prefill": true,
@ -10864,14 +10860,14 @@
"supports_tool_choice": true
},
"databricks/databricks-gemini-2-5-flash": {
"input_cost_per_token": 3.0001999999999996e-07,
"input_dbu_cost_per_token": 4.285999999999999e-06,
"input_cost_per_token": 3.0002e-07,
"input_dbu_cost_per_token": 4.286e-06,
"litellm_provider": "databricks",
"max_input_tokens": 1048576,
"max_output_tokens": 65535,
"max_tokens": 65535,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 2.49998e-06,
@ -10888,24 +10884,24 @@
"max_output_tokens": 65536,
"max_tokens": 65536,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 9.999990000000002e-06,
"output_cost_per_token": 9.99999e-06,
"output_dbu_cost_per_token": 0.000142857,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving",
"supports_function_calling": true,
"supports_tool_choice": true
},
"databricks/databricks-gemma-3-12b": {
"input_cost_per_token": 1.5000999999999998e-07,
"input_dbu_cost_per_token": 2.1429999999999996e-06,
"input_cost_per_token": 1.5001e-07,
"input_dbu_cost_per_token": 2.143e-06,
"litellm_provider": "databricks",
"max_input_tokens": 128000,
"max_output_tokens": 32000,
"max_tokens": 32000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 5.0001e-07,
@ -10920,10 +10916,10 @@
"max_output_tokens": 128000,
"max_tokens": 128000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 9.999990000000002e-06,
"output_cost_per_token": 9.99999e-06,
"output_dbu_cost_per_token": 0.000142857,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving"
},
@ -10935,25 +10931,25 @@
"max_output_tokens": 128000,
"max_tokens": 128000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 9.999990000000002e-06,
"output_cost_per_token": 9.99999e-06,
"output_dbu_cost_per_token": 0.000142857,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving"
},
"databricks/databricks-gpt-5-mini": {
"input_cost_per_token": 2.4997000000000006e-07,
"input_cost_per_token": 2.4997e-07,
"input_dbu_cost_per_token": 3.571e-06,
"litellm_provider": "databricks",
"max_input_tokens": 272000,
"max_output_tokens": 128000,
"max_tokens": 128000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 1.9999700000000004e-06,
"output_cost_per_token": 1.99997e-06,
"output_dbu_cost_per_token": 2.8571e-05,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving"
},
@ -10965,22 +10961,22 @@
"max_output_tokens": 128000,
"max_tokens": 128000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 3.9998000000000007e-07,
"output_dbu_cost_per_token": 5.714000000000001e-06,
"output_cost_per_token": 3.9998e-07,
"output_dbu_cost_per_token": 5.714e-06,
"source": "https://www.databricks.com/product/pricing/proprietary-foundation-model-serving"
},
"databricks/databricks-gpt-oss-120b": {
"input_cost_per_token": 1.5000999999999998e-07,
"input_dbu_cost_per_token": 2.1429999999999996e-06,
"input_cost_per_token": 1.5001e-07,
"input_dbu_cost_per_token": 2.143e-06,
"litellm_provider": "databricks",
"max_input_tokens": 131072,
"max_output_tokens": 131072,
"max_tokens": 131072,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 5.9997e-07,
@ -10995,15 +10991,15 @@
"max_output_tokens": 131072,
"max_tokens": 131072,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 3.0001999999999996e-07,
"output_dbu_cost_per_token": 4.285999999999999e-06,
"output_cost_per_token": 3.0002e-07,
"output_dbu_cost_per_token": 4.286e-06,
"source": "https://www.databricks.com/product/pricing/foundation-model-serving"
},
"databricks/databricks-gte-large-en": {
"input_cost_per_token": 1.2999000000000001e-07,
"input_cost_per_token": 1.2999e-07,
"input_dbu_cost_per_token": 1.857e-06,
"litellm_provider": "databricks",
"max_input_tokens": 8192,
@ -11028,7 +11024,7 @@
"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 1.5000300000000002e-06,
"output_cost_per_token": 1.50003e-06,
"output_dbu_cost_per_token": 2.1429e-05,
"source": "https://www.databricks.com/product/pricing/foundation-model-serving",
"supports_tool_choice": true
@ -11044,7 +11040,7 @@
"notes": "Databricks documentation now provides both DBU costs (_dbu_cost_per_token) and dollar costs(_cost_per_token)."
},
"mode": "chat",
"output_cost_per_token": 1.5000300000000002e-06,
"output_cost_per_token": 1.50003e-06,
"output_dbu_cost_per_token": 2.1429e-05,
"source": "https://www.databricks.com/product/pricing/foundation-model-serving",
"supports_tool_choice": true
@ -11060,24 +11056,24 @@
"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 1.5000020000000002e-05,
"output_cost_per_token": 1.500002e-05,
"output_dbu_cost_per_token": 0.000214286,
"source": "https://www.databricks.com/product/pricing/foundation-model-serving",
"supports_tool_choice": true
},
"databricks/databricks-meta-llama-3-1-8b-instruct": {
"input_cost_per_token": 1.5000999999999998e-07,
"input_dbu_cost_per_token": 2.1429999999999996e-06,
"input_cost_per_token": 1.5001e-07,
"input_dbu_cost_per_token": 2.143e-06,
"litellm_provider": "databricks",
"max_input_tokens": 200000,
"max_output_tokens": 128000,
"max_tokens": 128000,
"metadata": {
"notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
"notes": "Input/output cost per token is dbu cost * $0.070 Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 4.5003000000000007e-07,
"output_dbu_cost_per_token": 6.429000000000001e-06,
"output_cost_per_token": 4.5003e-07,
"output_dbu_cost_per_token": 6.429e-06,
"source": "https://www.databricks.com/product/pricing/foundation-model-serving"
},
"databricks/databricks-meta-llama-3-3-70b-instruct": {
@ -11091,7 +11087,7 @@
"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 1.5000300000000002e-06,
"output_cost_per_token": 1.50003e-06,
"output_dbu_cost_per_token": 2.1429e-05,
"source": "https://www.databricks.com/product/pricing/foundation-model-serving",
"supports_tool_choice": true
@ -11107,7 +11103,7 @@
"notes": "Input/output cost per token is dbu cost * $0.070, based on databricks Llama 3.1 70B conversion. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
},
"mode": "chat",
"output_cost_per_token": 2.9999900000000002e-06,
"output_cost_per_token": 2.99999e-06,
"output_dbu_cost_per_token": 4.2857e-05,
"source": "https://www.databricks.com/product/pricing/foundation-model-serving",
"supports_tool_choice": true
@ -16841,7 +16837,7 @@
"max_video_length": 1,
"max_videos_per_prompt": 10,
"metadata": {
"notes": "Rate limits not documented for gemini-exp-1114. Assuming same as gemini-1.5-pro.",
"notes": "Rate limits not documented for gemini-exp-1114 Assuming same as gemini-1.5-pro.",
"supports_tool_choice": true
},
"mode": "chat",
@ -16870,7 +16866,7 @@
"max_video_length": 1,
"max_videos_per_prompt": 10,
"metadata": {
"notes": "Rate limits not documented for gemini-exp-1206. Assuming same as gemini-1.5-pro.",
"notes": "Rate limits not documented for gemini-exp-1206 Assuming same as gemini-1.5-pro.",
"supports_tool_choice": true
},
"mode": "chat",
@ -22778,14 +22774,13 @@
},
"minimax.minimax-m2.5": {
"input_cost_per_token": 3e-07,
"output_cost_per_token": 1.2e-06,
"litellm_provider": "bedrock_converse",
"max_input_tokens": 1000000,
"max_output_tokens": 8192,
"max_tokens": 8192,
"mode": "chat",
"output_cost_per_token": 1.2e-06,
"supports_function_calling": true,
"supports_reasoning": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"source": "https://aws.amazon.com/bedrock/pricing/"
@ -34844,17 +34839,17 @@
},
"zai.glm-5": {
"input_cost_per_token": 1e-06,
"output_cost_per_token": 3.2e-06,
"litellm_provider": "bedrock_converse",
"max_input_tokens": 200000,
"max_output_tokens": 128000,
"max_tokens": 128000,
"mode": "chat",
"output_cost_per_token": 3.2e-06,
"source": "https://aws.amazon.com/bedrock/pricing/",
"supports_function_calling": true,
"supports_reasoning": true,
"supports_system_messages": true,
"supports_tool_choice": true
"supports_tool_choice": true,
"source": "https://aws.amazon.com/bedrock/pricing/"
},
"zai/glm-5": {
"cache_creation_input_token_cost": 0,
@ -39637,20 +39632,6 @@
}
]
},
"zai.glm-5": {
"input_cost_per_token": 1e-06,
"output_cost_per_token": 3.2e-06,
"litellm_provider": "bedrock_converse",
"max_input_tokens": 200000,
"max_output_tokens": 128000,
"max_tokens": 128000,
"mode": "chat",
"supports_function_calling": true,
"supports_reasoning": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"source": "https://aws.amazon.com/bedrock/pricing/"
},
"bedrock/us-east-1/zai.glm-5": {
"input_cost_per_token": 1e-06,
"output_cost_per_token": 3.2e-06,
@ -39679,45 +39660,6 @@
"supports_tool_choice": true,
"source": "https://aws.amazon.com/bedrock/pricing/"
},
"minimax.minimax-m2.5": {
"input_cost_per_token": 3e-07,
"output_cost_per_token": 1.2e-06,
"litellm_provider": "bedrock_converse",
"max_input_tokens": 1000000,
"max_output_tokens": 8192,
"max_tokens": 8192,
"mode": "chat",
"supports_function_calling": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"source": "https://aws.amazon.com/bedrock/pricing/"
},
"bedrock/us-east-1/minimax.minimax-m2.5": {
"input_cost_per_token": 3e-07,
"output_cost_per_token": 1.2e-06,
"litellm_provider": "bedrock",
"max_input_tokens": 1000000,
"max_output_tokens": 8192,
"max_tokens": 8192,
"mode": "chat",
"supports_function_calling": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"source": "https://aws.amazon.com/bedrock/pricing/"
},
"bedrock/us-west-2/minimax.minimax-m2.5": {
"input_cost_per_token": 3e-07,
"output_cost_per_token": 1.2e-06,
"litellm_provider": "bedrock",
"max_input_tokens": 1000000,
"max_output_tokens": 8192,
"max_tokens": 8192,
"mode": "chat",
"supports_function_calling": true,
"supports_system_messages": true,
"supports_tool_choice": true,
"source": "https://aws.amazon.com/bedrock/pricing/"
},
"bedrock/us-gov-east-1/anthropic.claude-haiku-4-5-20251001-v1:0": {
"cache_creation_input_token_cost": 1.5e-06,
"cache_read_input_token_cost": 1.2e-07,
@ -39784,7 +39726,7 @@
"max_input_tokens": 198000,
"max_output_tokens": 32000,
"input_cost_per_token": 1e-06,
"output_cost_per_token": 3.2000000000000003e-06,
"output_cost_per_token": 3.2e-06,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -39829,7 +39771,7 @@
"max_input_tokens": 200000,
"max_output_tokens": 24000,
"input_cost_per_token": 1.4e-07,
"output_cost_per_token": 8.000000000000001e-07,
"output_cost_per_token": 8e-07,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -39888,8 +39830,8 @@
"max_tokens": 8192,
"max_input_tokens": 32000,
"max_output_tokens": 8192,
"input_cost_per_token": 2.0000000000000002e-07,
"output_cost_per_token": 9.000000000000001e-07,
"input_cost_per_token": 2e-07,
"output_cost_per_token": 9e-07,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -39903,8 +39845,8 @@
"max_tokens": 8192,
"max_input_tokens": 128000,
"max_output_tokens": 8192,
"input_cost_per_token": 2.0000000000000002e-07,
"output_cost_per_token": 9.000000000000001e-07,
"input_cost_per_token": 2e-07,
"output_cost_per_token": 9e-07,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -39963,7 +39905,7 @@
"max_tokens": 32768,
"max_input_tokens": 256000,
"max_output_tokens": 32768,
"input_cost_per_token": 1.0000000000000001e-07,
"input_cost_per_token": 1e-07,
"output_cost_per_token": 1.5e-07,
"litellm_provider": "venice",
"mode": "chat",
@ -40008,7 +39950,7 @@
"max_tokens": 16384,
"max_input_tokens": 128000,
"max_output_tokens": 16384,
"input_cost_per_token": 4.5000000000000003e-07,
"input_cost_per_token": 4.5e-07,
"output_cost_per_token": 3.5e-06,
"litellm_provider": "venice",
"mode": "chat",
@ -40039,7 +39981,7 @@
"max_input_tokens": 256000,
"max_output_tokens": 16384,
"input_cost_per_token": 3.5e-07,
"output_cost_per_token": 1.8999999999999998e-06,
"output_cost_per_token": 1.9e-06,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -40144,7 +40086,7 @@
"max_input_tokens": 198000,
"max_output_tokens": 16384,
"input_cost_per_token": 1.2e-07,
"output_cost_per_token": 2.0000000000000002e-07,
"output_cost_per_token": 2e-07,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -40173,8 +40115,8 @@
"max_tokens": 30000,
"max_input_tokens": 1000000,
"max_output_tokens": 30000,
"input_cost_per_token": 2.3000000000000002e-07,
"output_cost_per_token": 5.699999999999999e-07,
"input_cost_per_token": 2.3e-07,
"output_cost_per_token": 5.7e-07,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -40368,7 +40310,7 @@
"max_tokens": 64000,
"max_input_tokens": 1000000,
"max_output_tokens": 64000,
"input_cost_per_token": 3.6000000000000003e-06,
"input_cost_per_token": 3.6e-06,
"output_cost_per_token": 1.8e-05,
"litellm_provider": "venice",
"mode": "chat",
@ -40429,7 +40371,7 @@
"max_input_tokens": 256000,
"max_output_tokens": 65536,
"input_cost_per_token": 7.5e-07,
"output_cost_per_token": 3.2000000000000003e-06,
"output_cost_per_token": 3.2e-06,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -40459,7 +40401,7 @@
"max_input_tokens": 1000000,
"max_output_tokens": 32768,
"input_cost_per_token": 1.73e-06,
"output_cost_per_token": 3.7959999999999997e-06,
"output_cost_per_token": 3.796e-06,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -40473,7 +40415,7 @@
"max_tokens": 32768,
"max_input_tokens": 1000000,
"max_output_tokens": 32768,
"input_cost_per_token": 1.7000000000000001e-07,
"input_cost_per_token": 1.7e-07,
"output_cost_per_token": 3.5e-07,
"litellm_provider": "venice",
"mode": "chat",
@ -40548,7 +40490,7 @@
"max_tokens": 65536,
"max_input_tokens": 256000,
"max_output_tokens": 65536,
"input_cost_per_token": 2.1899999999999998e-06,
"input_cost_per_token": 2.19e-06,
"output_cost_per_token": 1.75e-05,
"litellm_provider": "venice",
"mode": "chat",
@ -40563,7 +40505,7 @@
"max_tokens": 65536,
"max_input_tokens": 256000,
"max_output_tokens": 65536,
"input_cost_per_token": 2.1899999999999998e-06,
"input_cost_per_token": 2.19e-06,
"output_cost_per_token": 1.75e-05,
"litellm_provider": "venice",
"mode": "chat",
@ -40578,7 +40520,7 @@
"max_tokens": 128000,
"max_input_tokens": 400000,
"max_output_tokens": 128000,
"input_cost_per_token": 2.1899999999999998e-06,
"input_cost_per_token": 2.19e-06,
"output_cost_per_token": 1.75e-05,
"litellm_provider": "venice",
"mode": "chat",
@ -40698,7 +40640,7 @@
"max_tokens": 32768,
"max_input_tokens": 198000,
"max_output_tokens": 32768,
"input_cost_per_token": 3.4000000000000003e-07,
"input_cost_per_token": 3.4e-07,
"output_cost_per_token": 1.19e-06,
"litellm_provider": "venice",
"mode": "chat",
@ -40759,7 +40701,7 @@
"max_input_tokens": 256000,
"max_output_tokens": 32768,
"input_cost_per_token": 1.4e-07,
"output_cost_per_token": 8.000000000000001e-07,
"output_cost_per_token": 8e-07,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -40833,7 +40775,7 @@
"max_tokens": 32768,
"max_input_tokens": 128000,
"max_output_tokens": 32768,
"input_cost_per_token": 5.0000000000000004e-08,
"input_cost_per_token": 5e-08,
"output_cost_per_token": 1.9e-07,
"litellm_provider": "venice",
"mode": "chat",
@ -40863,7 +40805,7 @@
"max_tokens": 4096,
"max_input_tokens": 32000,
"max_output_tokens": 4096,
"input_cost_per_token": 5.0000000000000004e-08,
"input_cost_per_token": 5e-08,
"output_cost_per_token": 1.3e-07,
"litellm_provider": "venice",
"mode": "chat",
@ -40894,7 +40836,7 @@
"max_input_tokens": 128000,
"max_output_tokens": 4096,
"input_cost_per_token": 2.5e-07,
"output_cost_per_token": 9.000000000000001e-07,
"output_cost_per_token": 9e-07,
"litellm_provider": "venice",
"mode": "chat",
"supports_audio_input": false,
@ -40949,4 +40891,4 @@
"supports_vision": true,
"supports_web_search": true
}
}
}