[Feat] Add litellm.supports_reasoning() util to track if an llm supports reasoning (#9923)

* add supports_reasoning for xai models

* add "supports_reasoning": true for o1 series models

* add supports_reasoning util

* add litellm.supports_reasoning

* add supports reasoning for claude 3-7 models

* add deepseek as supports reasoning

* test_supports_reasoning

* add supports reasoning to model group info

* add supports_reasoning

* docs supports reasoning

* fix supports_reasoning test

* "supports_reasoning": false,

* fix test

* supports_reasoning

litellm/utils.py

@@ -484,7 +484,7 @@ def load_credentials_from_list(kwargs: dict):
 
 
 def get_dynamic_callbacks(
-    dynamic_callbacks: Optional[List[Union[str, Callable, CustomLogger]]]
+    dynamic_callbacks: Optional[List[Union[str, Callable, CustomLogger]]],
 ) -> List:
     returned_callbacks = litellm.callbacks.copy()
     if dynamic_callbacks:
@@ -516,9 +516,9 @@ def function_setup(  # noqa: PLR0915
         function_id: Optional[str] = kwargs["id"] if "id" in kwargs else None
 
         ## DYNAMIC CALLBACKS ##
-        dynamic_callbacks: Optional[
-            List[Union[str, Callable, CustomLogger]]
-        ] = kwargs.pop("callbacks", None)
+        dynamic_callbacks: Optional[List[Union[str, Callable, CustomLogger]]] = (
+            kwargs.pop("callbacks", None)
+        )
         all_callbacks = get_dynamic_callbacks(dynamic_callbacks=dynamic_callbacks)
 
         if len(all_callbacks) > 0:
@@ -1202,9 +1202,9 @@ def client(original_function):  # noqa: PLR0915
                         exception=e,
                         retry_policy=kwargs.get("retry_policy"),
                     )
-                    kwargs[
-                        "retry_policy"
-                    ] = reset_retry_policy()  # prevent infinite loops
+                    kwargs["retry_policy"] = (
+                        reset_retry_policy()
+                    )  # prevent infinite loops
                 litellm.num_retries = (
                     None  # set retries to None to prevent infinite loops
                 )
@@ -2229,6 +2229,15 @@ def supports_vision(model: str, custom_llm_provider: Optional[str] = None) -> bo
     )
 
 
+def supports_reasoning(model: str, custom_llm_provider: Optional[str] = None) -> bool:
+    """
+    Check if the given model supports reasoning and return a boolean value.
+    """
+    return _supports_factory(
+        model=model, custom_llm_provider=custom_llm_provider, key="supports_reasoning"
+    )
+
+
 def supports_embedding_image_input(
     model: str, custom_llm_provider: Optional[str] = None
 ) -> bool:
@@ -3004,16 +3013,16 @@ def get_optional_params(  # noqa: PLR0915
                     True  # so that main.py adds the function call to the prompt
                 )
                 if "tools" in non_default_params:
-                    optional_params[
-                        "functions_unsupported_model"
-                    ] = non_default_params.pop("tools")
+                    optional_params["functions_unsupported_model"] = (
+                        non_default_params.pop("tools")
+                    )
                     non_default_params.pop(
                         "tool_choice", None
                     )  # causes ollama requests to hang
                 elif "functions" in non_default_params:
-                    optional_params[
-                        "functions_unsupported_model"
-                    ] = non_default_params.pop("functions")
+                    optional_params["functions_unsupported_model"] = (
+                        non_default_params.pop("functions")
+                    )
             elif (
                 litellm.add_function_to_prompt
             ):  # if user opts to add it to prompt instead
@@ -3036,10 +3045,10 @@ def get_optional_params(  # noqa: PLR0915
 
     if "response_format" in non_default_params:
         if provider_config is not None:
-            non_default_params[
-                "response_format"
-            ] = provider_config.get_json_schema_from_pydantic_object(
-                response_format=non_default_params["response_format"]
+            non_default_params["response_format"] = (
+                provider_config.get_json_schema_from_pydantic_object(
+                    response_format=non_default_params["response_format"]
+                )
             )
         else:
             non_default_params["response_format"] = type_to_response_format_param(
@@ -4055,9 +4064,9 @@ def _count_characters(text: str) -> int:
 
 
 def get_response_string(response_obj: Union[ModelResponse, ModelResponseStream]) -> str:
-    _choices: Union[
-        List[Union[Choices, StreamingChoices]], List[StreamingChoices]
-    ] = response_obj.choices
+    _choices: Union[List[Union[Choices, StreamingChoices]], List[StreamingChoices]] = (
+        response_obj.choices
+    )
 
     response_str = ""
     for choice in _choices:
@@ -4597,6 +4606,7 @@ def _get_model_info_helper(  # noqa: PLR0915
                     "supports_native_streaming", None
                 ),
                 supports_web_search=_model_info.get("supports_web_search", False),
+                supports_reasoning=_model_info.get("supports_reasoning", False),
                 search_context_cost_per_query=_model_info.get(
                     "search_context_cost_per_query", None
                 ),
@@ -4669,6 +4679,7 @@ def get_model_info(model: str, custom_llm_provider: Optional[str] = None) -> Mod
             supports_audio_output: Optional[bool]
             supports_pdf_input: Optional[bool]
             supports_web_search: Optional[bool]
+            supports_reasoning: Optional[bool]
     Raises:
         Exception: If the model is not mapped yet.
 
@@ -6188,7 +6199,7 @@ def validate_chat_completion_user_messages(messages: List[AllMessageValues]):
 
 
 def validate_chat_completion_tool_choice(
-    tool_choice: Optional[Union[dict, str]]
+    tool_choice: Optional[Union[dict, str]],
 ) -> Optional[Union[dict, str]]:
     """
     Confirm the tool choice is passed in the OpenAI format.