Fix _supports_reasoning_effort_level for responses bridge

litellm/main.py

@@ -1356,6 +1356,13 @@ def completion(  # type: ignore # noqa: PLR0915
             api_key=api_key,
         )
 
+        ## RESPONSES API BRIDGE LOGIC ## - check early and normalize model name
+        responses_api_model_info, model = responses_api_bridge_check(
+            model=model,
+            custom_llm_provider=custom_llm_provider,
+            web_search_options=web_search_options,
+        )
+
         if not _should_allow_input_examples(
             custom_llm_provider=custom_llm_provider, model=model
         ):
@@ -1591,14 +1598,8 @@ def completion(  # type: ignore # noqa: PLR0915
                 timeout=timeout,
             )
 
-        ## RESPONSES API BRIDGE LOGIC ## - check if model has 'mode: responses' in litellm.model_cost map
-        model_info, model = responses_api_bridge_check(
-            model=model,
-            custom_llm_provider=custom_llm_provider,
-            web_search_options=web_search_options,
-        )
 
-        if model_info.get("mode") == "responses":
+        if responses_api_model_info.get("mode") == "responses":
             from litellm.completion_extras import responses_api_bridge
 
             return responses_api_bridge.completion(

tests/llm_translation/test_openai.py

@@ -1454,3 +1454,29 @@ def test_gpt_5_web_search():
 
     for chunk in response:
         print("chunk: ", chunk)
+
+
+def test_responses_gpt54_with_xhigh_reasoning():
+    """
+    Ensure chat->responses bridge sends the correct request payload for
+    openai/responses/gpt-5.4 with reasoning_effort="xhigh".
+    """
+    with patch("litellm.responses") as mock_responses:
+        # Stop execution right after request generation to avoid external API calls.
+        mock_responses.side_effect = RuntimeError("stop_after_request_build")
+
+        with pytest.raises(Exception):
+            litellm.completion(
+                model="openai/responses/gpt-5.4",
+                messages=[{"role": "user", "content": "What is 2+2?"}],
+                reasoning_effort="xhigh",
+                max_tokens=100,
+            )
+
+        mock_responses.assert_called_once()
+        request_body = mock_responses.call_args.kwargs
+
+        # The responses prefix should be stripped before routing.
+        assert request_body["model"] == "gpt-5.4"
+        # chat-completions reasoning_effort must map to Responses API reasoning.
+        assert request_body["reasoning"] == {"effort": "xhigh"}