[Fix] LiteLLM does not support new web_search tool (Responses API) (#14083)

* test_basic_openai_responses_with_websearch

* fix: ResponsesAPIResponse

* fix: StandardBuiltInToolCostTracking

litellm/litellm_core_utils/llm_cost_calc/tool_call_cost_tracking.py

@@ -580,7 +580,9 @@ class StandardBuiltInToolCostTracking:
             return WebSearchOptions(**kwargs.get("web_search_options", {}))
 
         tools = StandardBuiltInToolCostTracking._get_tools_from_kwargs(
-            kwargs, "web_search_preview"
+            kwargs=kwargs, tool_type="web_search_preview"
+        ) or StandardBuiltInToolCostTracking._get_tools_from_kwargs(
+            kwargs=kwargs, tool_type="web_search"
         )
         if tools:
             # Look for web search tool in the tools array
@@ -612,6 +614,8 @@ class StandardBuiltInToolCostTracking:
     def _is_web_search_tool_call(tool: Dict) -> bool:
         if tool.get("type", None) == "web_search_preview":
             return True
+        if tool.get("type", None) == "web_search":
+            return True
         if "search_context_size" in tool:
             return True
         return False

litellm/types/llms/openai.py

@@ -1038,7 +1038,7 @@ class ResponsesAPIResponse(BaseLiteLLMOpenAIResponseObject):
     parallel_tool_calls: bool
     temperature: Optional[float]
     tool_choice: ToolChoice
-    tools: Union[List[Tool], List[ResponseFunctionToolCall]]
+    tools: Union[List[Tool], List[ResponseFunctionToolCall], List[Dict[str, Any]]]
     top_p: Optional[float]
     max_output_tokens: Optional[int]
     previous_response_id: Optional[str]

tests/llm_responses_api_testing/test_openai_responses_api.py

@@ -1483,3 +1483,28 @@ async def test_openai_gpt5_reasoning_effort_parameter():
         # Validate the response
         print("Response:", json.dumps(response, indent=4, default=str))
 
+
+
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("stream", [True, False])
+async def test_basic_openai_responses_with_websearch(stream):
+    litellm._turn_on_debug()
+    request_model = "gpt-4o"
+    response = await litellm.aresponses(
+        model=request_model,
+        stream=stream,
+        input="hi",
+        tools=[
+            {
+                "type": "web_search",
+                "search_context_size": "low"
+            }
+        ]
+    )
+    if stream:
+        async for chunk in response:
+            print("chunk=", json.dumps(chunk, indent=4, default=str))
+    else:
+        print("response=", json.dumps(response, indent=4, default=str))