scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-10 03:28:53 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 85

fix(model_prices): carry anthropic behavior flags on deepinfra claude entries, move retired together models to deprecated list

Browse source 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
This commit is contained in:
Devin AI 2026-08-26 17:18:51 +00:00
parent d3ede97189
commit 07c9812739
3 changed files with 102 additions and 70 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

84
litellm/model_prices_and_context_window_backup.json
View file

@ -52689,18 +52689,21 @@
"source": "https://deepinfra.com/pricing"
},
"deepinfra/anthropic/claude-opus-4-8": {
"max_tokens": 1000000,
"max_input_tokens": 1000000,
"input_cost_per_token": 5e-06,
"output_cost_per_token": 2.5e-05,
"litellm_provider": "deepinfra",
"max_input_tokens": 1000000,
"max_tokens": 1000000,
"mode": "chat",
"output_cost_per_token": 2.5e-05,
"prompt_cache_min_tokens": 1024,
"source": "https://deepinfra.com/pricing",
"supports_adaptive_thinking": true,
"supports_function_calling": true,
"supports_tool_choice": true,
"supports_response_schema": true,
"supports_reasoning": true,
"supports_vision": true,
"source": "https://deepinfra.com/pricing"
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true
},
"deepinfra/anthropic/claude-sonnet-4-6": {
"max_tokens": 1000000,
@ -52890,18 +52893,21 @@
"source": "https://deepinfra.com/pricing"
},
"deepinfra/anthropic/claude-opus-5": {
"max_tokens": 1000000,
"max_input_tokens": 1000000,
"input_cost_per_token": 5e-06,
"output_cost_per_token": 2.5e-05,
"litellm_provider": "deepinfra",
"max_input_tokens": 1000000,
"max_tokens": 1000000,
"mode": "chat",
"output_cost_per_token": 2.5e-05,
"prompt_cache_min_tokens": 512,
"source": "https://deepinfra.com/pricing",
"supports_adaptive_thinking": true,
"supports_function_calling": true,
"supports_tool_choice": true,
"supports_response_schema": true,
"supports_reasoning": true,
"supports_vision": true,
"source": "https://deepinfra.com/pricing"
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true
},
"deepinfra/thinkingmachines/Inkling": {
"max_tokens": 524288,
@ -53199,18 +53205,21 @@
"source": "https://deepinfra.com/pricing"
},
"deepinfra/anthropic/claude-sonnet-5": {
"max_tokens": 1000000,
"max_input_tokens": 1000000,
"input_cost_per_token": 2e-06,
"output_cost_per_token": 1e-05,
"litellm_provider": "deepinfra",
"max_input_tokens": 1000000,
"max_tokens": 1000000,
"mode": "chat",
"output_cost_per_token": 1e-05,
"prompt_cache_min_tokens": 1024,
"source": "https://deepinfra.com/pricing",
"supports_adaptive_thinking": true,
"supports_function_calling": true,
"supports_tool_choice": true,
"supports_response_schema": true,
"supports_reasoning": true,
"supports_vision": true,
"source": "https://deepinfra.com/pricing"
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true
},
"deepinfra/Qwen/Qwen3.5-397B-A17B": {
"max_tokens": 262144,
@ -53288,18 +53297,22 @@
"source": "https://deepinfra.com/pricing"
},
"deepinfra/anthropic/claude-fable-5": {
"max_tokens": 1000000,
"max_input_tokens": 1000000,
"input_cost_per_token": 1e-05,
"output_cost_per_token": 5e-05,
"litellm_provider": "deepinfra",
"max_input_tokens": 1000000,
"max_tokens": 1000000,
"mode": "chat",
"output_cost_per_token": 5e-05,
"prompt_cache_min_tokens": 512,
"source": "https://deepinfra.com/pricing",
"supports_adaptive_thinking": true,
"supports_function_calling": true,
"supports_tool_choice": true,
"supports_response_schema": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true,
"source": "https://deepinfra.com/pricing"
"thinking_always_on": true
},
"deepinfra/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B": {
"max_tokens": 262144,
@ -53408,18 +53421,21 @@
"source": "https://deepinfra.com/pricing"
},
"deepinfra/anthropic/claude-opus-4-7": {
"max_tokens": 1000000,
"max_input_tokens": 1000000,
"input_cost_per_token": 5e-06,
"output_cost_per_token": 2.5e-05,
"litellm_provider": "deepinfra",
"max_input_tokens": 1000000,
"max_tokens": 1000000,
"mode": "chat",
"output_cost_per_token": 2.5e-05,
"prompt_cache_min_tokens": 2048,
"source": "https://deepinfra.com/pricing",
"supports_adaptive_thinking": true,
"supports_function_calling": true,
"supports_tool_choice": true,
"supports_response_schema": true,
"supports_reasoning": true,
"supports_vision": true,
"source": "https://deepinfra.com/pricing"
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true
},
"deepinfra/Qwen/Qwen3.6-27B": {
"max_tokens": 262144,

84
model_prices_and_context_window.json
View file

@ -52689,18 +52689,21 @@
"source": "https://deepinfra.com/pricing"
},
"deepinfra/anthropic/claude-opus-4-8": {
"max_tokens": 1000000,
"max_input_tokens": 1000000,
"input_cost_per_token": 5e-06,
"output_cost_per_token": 2.5e-05,
"litellm_provider": "deepinfra",
"max_input_tokens": 1000000,
"max_tokens": 1000000,
"mode": "chat",
"output_cost_per_token": 2.5e-05,
"prompt_cache_min_tokens": 1024,
"source": "https://deepinfra.com/pricing",
"supports_adaptive_thinking": true,
"supports_function_calling": true,
"supports_tool_choice": true,
"supports_response_schema": true,
"supports_reasoning": true,
"supports_vision": true,
"source": "https://deepinfra.com/pricing"
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true
},
"deepinfra/anthropic/claude-sonnet-4-6": {
"max_tokens": 1000000,
@ -52890,18 +52893,21 @@
"source": "https://deepinfra.com/pricing"
},
"deepinfra/anthropic/claude-opus-5": {
"max_tokens": 1000000,
"max_input_tokens": 1000000,
"input_cost_per_token": 5e-06,
"output_cost_per_token": 2.5e-05,
"litellm_provider": "deepinfra",
"max_input_tokens": 1000000,
"max_tokens": 1000000,
"mode": "chat",
"output_cost_per_token": 2.5e-05,
"prompt_cache_min_tokens": 512,
"source": "https://deepinfra.com/pricing",
"supports_adaptive_thinking": true,
"supports_function_calling": true,
"supports_tool_choice": true,
"supports_response_schema": true,
"supports_reasoning": true,
"supports_vision": true,
"source": "https://deepinfra.com/pricing"
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true
},
"deepinfra/thinkingmachines/Inkling": {
"max_tokens": 524288,
@ -53199,18 +53205,21 @@
"source": "https://deepinfra.com/pricing"
},
"deepinfra/anthropic/claude-sonnet-5": {
"max_tokens": 1000000,
"max_input_tokens": 1000000,
"input_cost_per_token": 2e-06,
"output_cost_per_token": 1e-05,
"litellm_provider": "deepinfra",
"max_input_tokens": 1000000,
"max_tokens": 1000000,
"mode": "chat",
"output_cost_per_token": 1e-05,
"prompt_cache_min_tokens": 1024,
"source": "https://deepinfra.com/pricing",
"supports_adaptive_thinking": true,
"supports_function_calling": true,
"supports_tool_choice": true,
"supports_response_schema": true,
"supports_reasoning": true,
"supports_vision": true,
"source": "https://deepinfra.com/pricing"
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true
},
"deepinfra/Qwen/Qwen3.5-397B-A17B": {
"max_tokens": 262144,
@ -53288,18 +53297,22 @@
"source": "https://deepinfra.com/pricing"
},
"deepinfra/anthropic/claude-fable-5": {
"max_tokens": 1000000,
"max_input_tokens": 1000000,
"input_cost_per_token": 1e-05,
"output_cost_per_token": 5e-05,
"litellm_provider": "deepinfra",
"max_input_tokens": 1000000,
"max_tokens": 1000000,
"mode": "chat",
"output_cost_per_token": 5e-05,
"prompt_cache_min_tokens": 512,
"source": "https://deepinfra.com/pricing",
"supports_adaptive_thinking": true,
"supports_function_calling": true,
"supports_tool_choice": true,
"supports_response_schema": true,
"supports_reasoning": true,
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true,
"source": "https://deepinfra.com/pricing"
"thinking_always_on": true
},
"deepinfra/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B": {
"max_tokens": 262144,
@ -53408,18 +53421,21 @@
"source": "https://deepinfra.com/pricing"
},
"deepinfra/anthropic/claude-opus-4-7": {
"max_tokens": 1000000,
"max_input_tokens": 1000000,
"input_cost_per_token": 5e-06,
"output_cost_per_token": 2.5e-05,
"litellm_provider": "deepinfra",
"max_input_tokens": 1000000,
"max_tokens": 1000000,
"mode": "chat",
"output_cost_per_token": 2.5e-05,
"prompt_cache_min_tokens": 2048,
"source": "https://deepinfra.com/pricing",
"supports_adaptive_thinking": true,
"supports_function_calling": true,
"supports_tool_choice": true,
"supports_response_schema": true,
"supports_reasoning": true,
"supports_vision": true,
"source": "https://deepinfra.com/pricing"
"supports_response_schema": true,
"supports_sampling_params": false,
"supports_tool_choice": true,
"supports_vision": true
},
"deepinfra/Qwen/Qwen3.6-27B": {
"max_tokens": 262144,

4
tests/test_litellm/test_together_ai_model_metadata.py
View file

@ -31,16 +31,16 @@ SERVERLESS_CHAT_MODELS: Final = (
"together_ai/meta-models/Muse-Glimmer-30B",
"together_ai/google/gemma-4-31B-it",
"together_ai/pearl-ai/gemma-4-31b-it",
"together_ai/google/gemma-3n-E4B-it",
"together_ai/arize-ai/qwen-2-1.5b-instruct",
"together_ai/Prism-ML/Ternary-Bonsai-27B",
"together_ai/meta-llama/Llama-Guard-4-12B",
"together_ai/openai/gpt-oss-120b",
"together_ai/openai/gpt-oss-20b",
"together_ai/meta-llama/Llama-3.3-70B-Instruct-Turbo",
)
DEPRECATED_MODELS: Final = {
"together_ai/google/gemma-3n-E4B-it": "2026-08-25",
"together_ai/meta-llama/Llama-Guard-4-12B": "2026-08-25",
"together_ai/Qwen/Qwen3-235B-A22B-Instruct-2507-tput": "2026-07-10",
"together_ai/Qwen/Qwen3.5-397B-A17B": "2026-06-29",
"together_ai/Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8": "2026-06-04",