fixes for various models with incorrect pricing

This commit is contained in:
Ishaan Jaffer
2025-10-17 11:51:22 -07:00
parent 8b179f0a89
commit 3bc8f76d12
2 changed files with 118 additions and 94 deletions
@@ -22863,13 +22863,13 @@
"mode": "chat"
},
"watsonx/ibm/granite-3-8b-instruct": {
"input_cost_per_token": 0.0002,
"input_cost_per_token": 0.2e-06,
"litellm_provider": "watsonx",
"max_input_tokens": 8192,
"max_output_tokens": 1024,
"max_tokens": 8192,
"mode": "chat",
"output_cost_per_token": 0.0002,
"output_cost_per_token": 0.2e-06,
"supports_audio_input": false,
"supports_audio_output": false,
"supports_function_calling": true,
@@ -22887,7 +22887,7 @@
"max_output_tokens": 16384,
"max_tokens": 131072,
"mode": "chat",
"output_cost_per_token": 1e-05,
"output_cost_per_token": 10e-06,
"supports_audio_input": false,
"supports_audio_output": false,
"supports_function_calling": true,
@@ -22926,8 +22926,8 @@
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.0001,
"output_cost_per_token": 0.00025,
"input_cost_per_token": 0.6e-06,
"output_cost_per_token": 0.6e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -22938,8 +22938,8 @@
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.0005,
"output_cost_per_token": 0.002,
"input_cost_per_token": 0.6e-06,
"output_cost_per_token": 0.6e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -22950,8 +22950,8 @@
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.0005,
"output_cost_per_token": 0.002,
"input_cost_per_token": 0.6e-06,
"output_cost_per_token": 0.6e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -22962,8 +22962,8 @@
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.00025,
"output_cost_per_token": 0.001,
"input_cost_per_token": 0.2e-06,
"output_cost_per_token": 0.2e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -22974,8 +22974,8 @@
"max_tokens": 20480,
"max_input_tokens": 20480,
"max_output_tokens": 20480,
"input_cost_per_token": 0.000625,
"output_cost_per_token": 0.0025,
"input_cost_per_token": 0.06e-06,
"output_cost_per_token": 0.25e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -22986,8 +22986,8 @@
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.00015,
"output_cost_per_token": 0.0006,
"input_cost_per_token": 0.1e-06,
"output_cost_per_token": 0.1e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -22998,8 +22998,8 @@
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.00025,
"output_cost_per_token": 0.001,
"input_cost_per_token": 0.2e-06,
"output_cost_per_token": 0.2e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -23010,8 +23010,8 @@
"max_tokens": 512,
"max_input_tokens": 512,
"max_output_tokens": 512,
"input_cost_per_token": 0.000625,
"output_cost_per_token": 0.000625,
"input_cost_per_token": 0.38e-06,
"output_cost_per_token": 0.38e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -23022,8 +23022,8 @@
"max_tokens": 512,
"max_input_tokens": 512,
"max_output_tokens": 512,
"input_cost_per_token": 0.000625,
"output_cost_per_token": 0.000625,
"input_cost_per_token": 0.38e-06,
"output_cost_per_token": 0.38e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -23034,8 +23034,8 @@
"max_tokens": 512,
"max_input_tokens": 512,
"max_output_tokens": 512,
"input_cost_per_token": 0.000625,
"output_cost_per_token": 0.000625,
"input_cost_per_token": 0.38e-06,
"output_cost_per_token": 0.38e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -23046,8 +23046,8 @@
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.00015,
"output_cost_per_token": 0.0006,
"input_cost_per_token": 0.1e-06,
"output_cost_per_token": 0.1e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -23058,8 +23058,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.00025,
"output_cost_per_token": 0.001,
"input_cost_per_token": 0.35e-06,
"output_cost_per_token": 0.35e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -23070,8 +23070,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.0001,
"output_cost_per_token": 0.0002,
"input_cost_per_token": 0.1e-06,
"output_cost_per_token": 0.1e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -23082,8 +23082,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.00015,
"output_cost_per_token": 0.0006,
"input_cost_per_token": 0.15e-06,
"output_cost_per_token": 0.15e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -23094,8 +23094,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.002,
"output_cost_per_token": 0.008,
"input_cost_per_token": 2e-06,
"output_cost_per_token": 2e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -23106,8 +23106,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.002,
"output_cost_per_token": 0.006,
"input_cost_per_token": 0.71e-06,
"output_cost_per_token": 0.71e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -23118,8 +23118,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.0005,
"output_cost_per_token": 0.002,
"input_cost_per_token": 0.35e-06,
"output_cost_per_token": 1.4e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -23130,8 +23130,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.00025,
"output_cost_per_token": 0.001,
"input_cost_per_token": 0.35e-06,
"output_cost_per_token": 0.35e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -23142,8 +23142,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.00225,
"output_cost_per_token": 0.00675,
"input_cost_per_token": 3e-06,
"output_cost_per_token": 10e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -23154,8 +23154,20 @@
"max_tokens": 32000,
"max_input_tokens": 32000,
"max_output_tokens": 32000,
"input_cost_per_token": 0.0002,
"output_cost_per_token": 0.0006,
"input_cost_per_token": 0.1e-06,
"output_cost_per_token": 0.3e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
"supports_parallel_function_calling": true,
"supports_vision": false
},
"watsonx/mistralai/mistral-small-3-1-24b-instruct-2503": {
"max_tokens": 32000,
"max_input_tokens": 32000,
"max_output_tokens": 32000,
"input_cost_per_token": 0.1e-06,
"output_cost_per_token": 0.3e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -23166,8 +23178,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.00015,
"output_cost_per_token": 0.00015,
"input_cost_per_token": 0.35e-06,
"output_cost_per_token": 0.35e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -23190,8 +23202,8 @@
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.0005,
"output_cost_per_token": 0.002,
"input_cost_per_token": 1.8e-06,
"output_cost_per_token": 1.8e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
+59 -47
View File
@@ -22863,13 +22863,13 @@
"mode": "chat"
},
"watsonx/ibm/granite-3-8b-instruct": {
"input_cost_per_token": 0.0002,
"input_cost_per_token": 0.2e-06,
"litellm_provider": "watsonx",
"max_input_tokens": 8192,
"max_output_tokens": 1024,
"max_tokens": 8192,
"mode": "chat",
"output_cost_per_token": 0.0002,
"output_cost_per_token": 0.2e-06,
"supports_audio_input": false,
"supports_audio_output": false,
"supports_function_calling": true,
@@ -22887,7 +22887,7 @@
"max_output_tokens": 16384,
"max_tokens": 131072,
"mode": "chat",
"output_cost_per_token": 1e-05,
"output_cost_per_token": 10e-06,
"supports_audio_input": false,
"supports_audio_output": false,
"supports_function_calling": true,
@@ -22926,8 +22926,8 @@
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.0001,
"output_cost_per_token": 0.00025,
"input_cost_per_token": 0.6e-06,
"output_cost_per_token": 0.6e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -22938,8 +22938,8 @@
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.0005,
"output_cost_per_token": 0.002,
"input_cost_per_token": 0.6e-06,
"output_cost_per_token": 0.6e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -22950,8 +22950,8 @@
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.0005,
"output_cost_per_token": 0.002,
"input_cost_per_token": 0.6e-06,
"output_cost_per_token": 0.6e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -22962,8 +22962,8 @@
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.00025,
"output_cost_per_token": 0.001,
"input_cost_per_token": 0.2e-06,
"output_cost_per_token": 0.2e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -22974,8 +22974,8 @@
"max_tokens": 20480,
"max_input_tokens": 20480,
"max_output_tokens": 20480,
"input_cost_per_token": 0.000625,
"output_cost_per_token": 0.0025,
"input_cost_per_token": 0.06e-06,
"output_cost_per_token": 0.25e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -22986,8 +22986,8 @@
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.00015,
"output_cost_per_token": 0.0006,
"input_cost_per_token": 0.1e-06,
"output_cost_per_token": 0.1e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -22998,8 +22998,8 @@
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.00025,
"output_cost_per_token": 0.001,
"input_cost_per_token": 0.2e-06,
"output_cost_per_token": 0.2e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -23010,8 +23010,8 @@
"max_tokens": 512,
"max_input_tokens": 512,
"max_output_tokens": 512,
"input_cost_per_token": 0.000625,
"output_cost_per_token": 0.000625,
"input_cost_per_token": 0.38e-06,
"output_cost_per_token": 0.38e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -23022,8 +23022,8 @@
"max_tokens": 512,
"max_input_tokens": 512,
"max_output_tokens": 512,
"input_cost_per_token": 0.000625,
"output_cost_per_token": 0.000625,
"input_cost_per_token": 0.38e-06,
"output_cost_per_token": 0.38e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -23034,8 +23034,8 @@
"max_tokens": 512,
"max_input_tokens": 512,
"max_output_tokens": 512,
"input_cost_per_token": 0.000625,
"output_cost_per_token": 0.000625,
"input_cost_per_token": 0.38e-06,
"output_cost_per_token": 0.38e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -23046,8 +23046,8 @@
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.00015,
"output_cost_per_token": 0.0006,
"input_cost_per_token": 0.1e-06,
"output_cost_per_token": 0.1e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -23058,8 +23058,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.00025,
"output_cost_per_token": 0.001,
"input_cost_per_token": 0.35e-06,
"output_cost_per_token": 0.35e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -23070,8 +23070,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.0001,
"output_cost_per_token": 0.0002,
"input_cost_per_token": 0.1e-06,
"output_cost_per_token": 0.1e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -23082,8 +23082,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.00015,
"output_cost_per_token": 0.0006,
"input_cost_per_token": 0.15e-06,
"output_cost_per_token": 0.15e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -23094,8 +23094,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.002,
"output_cost_per_token": 0.008,
"input_cost_per_token": 2e-06,
"output_cost_per_token": 2e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -23106,8 +23106,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.002,
"output_cost_per_token": 0.006,
"input_cost_per_token": 0.71e-06,
"output_cost_per_token": 0.71e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -23118,8 +23118,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.0005,
"output_cost_per_token": 0.002,
"input_cost_per_token": 0.35e-06,
"output_cost_per_token": 1.4e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -23130,8 +23130,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.00025,
"output_cost_per_token": 0.001,
"input_cost_per_token": 0.35e-06,
"output_cost_per_token": 0.35e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -23142,8 +23142,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.00225,
"output_cost_per_token": 0.00675,
"input_cost_per_token": 3e-06,
"output_cost_per_token": 10e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -23154,8 +23154,20 @@
"max_tokens": 32000,
"max_input_tokens": 32000,
"max_output_tokens": 32000,
"input_cost_per_token": 0.0002,
"output_cost_per_token": 0.0006,
"input_cost_per_token": 0.1e-06,
"output_cost_per_token": 0.3e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
"supports_parallel_function_calling": true,
"supports_vision": false
},
"watsonx/mistralai/mistral-small-3-1-24b-instruct-2503": {
"max_tokens": 32000,
"max_input_tokens": 32000,
"max_output_tokens": 32000,
"input_cost_per_token": 0.1e-06,
"output_cost_per_token": 0.3e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": true,
@@ -23166,8 +23178,8 @@
"max_tokens": 128000,
"max_input_tokens": 128000,
"max_output_tokens": 128000,
"input_cost_per_token": 0.00015,
"output_cost_per_token": 0.00015,
"input_cost_per_token": 0.35e-06,
"output_cost_per_token": 0.35e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,
@@ -23190,8 +23202,8 @@
"max_tokens": 8192,
"max_input_tokens": 8192,
"max_output_tokens": 8192,
"input_cost_per_token": 0.0005,
"output_cost_per_token": 0.002,
"input_cost_per_token": 1.8e-06,
"output_cost_per_token": 1.8e-06,
"litellm_provider": "watsonx",
"mode": "chat",
"supports_function_calling": false,