fix: OpenAI base_url default and reasoning effort model option.

mellea/backends/openai.py

@@ -631,15 +631,20 @@ async def _generate_from_chat_context_standard(
         formatted_tools = convert_tools_to_json(tools)
         use_tools = len(formatted_tools) > 0
 
+        # Build optional reasoning parameters
+        reasoning_params = {}
+        if thinking is not None:
+            reasoning_params["reasoning_effort"] = thinking
+
         chat_response: Coroutine[
             Any, Any, ChatCompletion | openai.AsyncStream[ChatCompletionChunk]
         ] = self._async_client.chat.completions.create(
             model=self._hf_model_id,
             messages=conversation,  # type: ignore
-            reasoning_effort=thinking,  # type: ignore
             tools=formatted_tools if use_tools else None,  # type: ignore
             # parallel_tool_calls=False, # We only support calling one tool per turn. But we do the choosing on our side so we leave this False.
             **extra_params,
+            **reasoning_params,  # type: ignore
             **self._make_backend_specific_and_remove(
                 model_opts, is_chat_context=ctx.is_chat_context
             ),

test/backends/test_openai_ollama.py

@@ -216,6 +216,46 @@ async def get_client_async():
     assert len(backend._client_cache.cache.values()) == 2
 
 
+async def test_reasoning_effort_conditional_passing(backend):
+    """Test that reasoning_effort is only passed to API when not None."""
+    from unittest.mock import AsyncMock, MagicMock, patch
+
+    ctx = ChatContext()
+    ctx = ctx.add(CBlock(value="Test"))
+
+    mock_response = MagicMock()
+    mock_response.choices = [MagicMock()]
+    mock_response.choices[0].message = MagicMock()
+    mock_response.choices[0].message.content = "Response"
+    mock_response.choices[0].message.role = "assistant"
+
+    # Test 1: reasoning_effort should NOT be passed when not specified
+    with patch.object(
+        backend._async_client.chat.completions, "create", new_callable=AsyncMock
+    ) as mock_create:
+        mock_create.return_value = mock_response
+        await backend.generate_from_chat_context(
+            CBlock(value="Hi"), ctx, model_options={}
+        )
+        call_kwargs = mock_create.call_args.kwargs
+        assert "reasoning_effort" not in call_kwargs, (
+            "reasoning_effort should not be passed when not specified"
+        )
+
+    # Test 2: reasoning_effort SHOULD be passed when specified
+    with patch.object(
+        backend._async_client.chat.completions, "create", new_callable=AsyncMock
+    ) as mock_create:
+        mock_create.return_value = mock_response
+        await backend.generate_from_chat_context(
+            CBlock(value="Hi"), ctx, model_options={ModelOption.THINKING: "medium"}
+        )
+        call_kwargs = mock_create.call_args.kwargs
+        assert call_kwargs.get("reasoning_effort") == "medium", (
+            "reasoning_effort should be passed with correct value when specified"
+        )
+
+
 if __name__ == "__main__":
     import pytest