Fix response_format validation for OpenAI-compatible models

- Allow OpenAI provider models to use response_format without validation errors
- Add comprehensive tests for OpenAI-compatible models like qwen-plus
- Ensure non-OpenAI providers still use original validation logic
- Fixes issue #3174 where structured output failed for third-party OpenAI-compatible models

Co-Authored-By: Jo\u00E3o <joao@crewai.com>

src/crewai/llm.py

@@ -1090,14 +1090,18 @@ class LLM(BaseLLM):
           - If no slash is present, "openai" is assumed.
         """
         provider = self._get_custom_llm_provider()
-        if self.response_format is not None and not supports_response_schema(
-            model=self.model,
-            custom_llm_provider=provider,
-        ):
-            raise ValueError(
-                f"The model {self.model} does not support response_format for provider '{provider}'. "
-                "Please remove response_format or use a supported model."
-            )
+        if self.response_format is not None:
+            if provider == "openai":
+                return
+            
+            if not supports_response_schema(
+                model=self.model,
+                custom_llm_provider=provider,
+            ):
+                raise ValueError(
+                    f"The model {self.model} does not support response_format for provider '{provider}'. "
+                    "Please remove response_format or use a supported model."
+                )
 
     def supports_function_calling(self) -> bool:
         try:

tests/llm_test.py

@@ -259,6 +259,41 @@ def test_validate_call_params_no_response_format():
     llm._validate_call_params()
 
 
+def test_openai_compatible_models_response_format():
+    """Test that OpenAI-compatible models support response_format without validation errors."""
+    from pydantic import BaseModel
+    
+    class TestResponse(BaseModel):
+        content: str
+    
+    llm = LLM(model="openai/qwen-plus", response_format=TestResponse)
+    llm._validate_call_params()
+    
+    llm = LLM(model="openai/custom-model", response_format=TestResponse)
+    llm._validate_call_params()
+    
+    llm = LLM(model="openai/gpt-4", response_format=TestResponse)
+    llm._validate_call_params()
+
+
+def test_non_openai_providers_still_use_validation():
+    """Test that non-OpenAI providers still use the original validation logic."""
+    from pydantic import BaseModel
+    
+    class TestResponse(BaseModel):
+        content: str
+    
+    with patch("crewai.llm.supports_response_schema", return_value=False):
+        llm = LLM(model="gemini/gemini-1.5-pro", response_format=TestResponse)
+        with pytest.raises(ValueError) as excinfo:
+            llm._validate_call_params()
+        assert "does not support response_format" in str(excinfo.value)
+    
+    with patch("crewai.llm.supports_response_schema", return_value=True):
+        llm = LLM(model="anthropic/claude-3", response_format=TestResponse)
+        llm._validate_call_params()
+
+
 @pytest.mark.vcr(filter_headers=["authorization"], filter_query_parameters=["key"])
 @pytest.mark.parametrize(
     "model",