fix(llms): honor reasoning-model params in OpenAIStructuredLLM (#5458)

Co-authored-by: ly-wang19 <ly-wang19@users.noreply.github.com>
Co-authored-by: kartik-mem0 <kartik.labhshetwar@mem0.ai>

mem0/llms/openai_structured.py

@@ -36,11 +36,8 @@ class OpenAIStructuredLLM(LLMBase):
         Returns:
             str: The generated response.
         """
-        params = {
+        params = self._get_supported_params(messages=messages)
-            "model": self.config.model,
+        params["model"] = self.config.model
-            "messages": messages,
-            "temperature": self.config.temperature,
-        }
 
         if response_format:
             params["response_format"] = response_format

tests/llms/test_openai_structured.py

@@ -0,0 +1,51 @@
+from unittest.mock import Mock, patch
+
+import pytest
+
+from mem0.configs.llms.openai import OpenAIConfig
+from mem0.llms.openai_structured import OpenAIStructuredLLM
+
+
+@pytest.fixture
+def mock_openai_client():
+    with patch("mem0.llms.openai_structured.OpenAI") as mock_openai:
+        mock_client = Mock()
+        mock_openai.return_value = mock_client
+        yield mock_client
+
+
+def _mock_parse(mock_client, content="ok"):
+    mock_response = Mock()
+    mock_response.choices = [Mock(message=Mock(content=content))]
+    mock_client.beta.chat.completions.parse.return_value = mock_response
+
+
+def test_reasoning_model_drops_temperature(mock_openai_client):
+    """Reasoning models reject `temperature`; structured output must not send it."""
+    config = OpenAIConfig(model="o3-mini", reasoning_effort="low")
+    llm = OpenAIStructuredLLM(config)
+    _mock_parse(mock_openai_client)
+
+    llm.generate_response([{"role": "user", "content": "Hello"}])
+
+    call_kwargs = mock_openai_client.beta.chat.completions.parse.call_args[1]
+    assert "temperature" not in call_kwargs  # reasoning models don't accept temperature
+    assert "max_tokens" not in call_kwargs  # also dropped for reasoning models
+    assert "top_p" not in call_kwargs  # also dropped for reasoning models
+    assert call_kwargs["reasoning_effort"] == "low"
+    assert call_kwargs["model"] == "o3-mini"
+
+
+def test_regular_model_sends_sampling_params(mock_openai_client):
+    """Regular models still receive the standard sampling params."""
+    config = OpenAIConfig(model="gpt-4o", temperature=0.3)
+    llm = OpenAIStructuredLLM(config)
+    _mock_parse(mock_openai_client)
+
+    llm.generate_response([{"role": "user", "content": "Hello"}])
+
+    call_kwargs = mock_openai_client.beta.chat.completions.parse.call_args[1]
+    assert call_kwargs["temperature"] == 0.3
+    assert "max_tokens" in call_kwargs  # standard sampling params still forwarded
+    assert "top_p" in call_kwargs
+    assert call_kwargs["model"] == "gpt-4o"