Fix routing of tool call + reasoning effor for gpt-5.4

litellm/main.py

@@ -99,6 +99,7 @@ from litellm.llms.base_llm.base_model_iterator import (
 from litellm.llms.bedrock.common_utils import BedrockModelInfo
 from litellm.llms.cohere.common_utils import CohereModelInfo
 from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, HTTPHandler
+from litellm.llms.openai.chat.gpt_5_transformation import OpenAIGPT5Config
 from litellm.llms.openai_like.json_loader import JSONProviderRegistry
 from litellm.llms.vertex_ai.common_utils import (
     VertexAIModelRoute,
@@ -934,6 +935,8 @@ def responses_api_bridge_check(
     model: str,
     custom_llm_provider: str,
     web_search_options: Optional[OpenAIWebSearchOptions] = None,
+    tools: Optional[List[Any]] = None,
+    reasoning_effort: Optional[Any] = None,
 ) -> Tuple[dict, str]:
     model_info: Dict[str, Any] = {}
     try:
@@ -951,6 +954,17 @@ def responses_api_bridge_check(
         if web_search_options is not None and custom_llm_provider == "xai":
             model_info["mode"] = "responses"
             model = model.replace("responses/", "")
+
+        # OpenAI gpt-5.4 chat-completions calls with both tools + reasoning_effort
+        # must be bridged to Responses API.
+        if (
+            custom_llm_provider == "openai"
+            and OpenAIGPT5Config.is_model_gpt_5_4_plus_model(model)
+            and tools
+            and reasoning_effort is not None
+        ):
+            model_info["mode"] = "responses"
+            model = model.replace("responses/", "")
     except Exception as e:
         verbose_logger.debug("Error getting model info: {}".format(e))
 
@@ -1596,6 +1610,8 @@ def completion(  # type: ignore # noqa: PLR0915
             model=model,
             custom_llm_provider=custom_llm_provider,
             web_search_options=web_search_options,
+            tools=tools,
+            reasoning_effort=reasoning_effort,
         )
 
         if model_info.get("mode") == "responses":

tests/test_litellm/test_main.py

@@ -627,6 +627,57 @@ def test_responses_api_bridge_check_gpt_5_4_pro():
         )
 
 
+def test_responses_api_bridge_check_gpt_5_4_tools_plus_reasoning_routes_to_responses():
+    """gpt-5.4 with both tools and reasoning_effort should route to Responses API."""
+    from litellm.main import responses_api_bridge_check
+
+    with patch("litellm.main._get_model_info_helper") as mock_get_model_info:
+        mock_get_model_info.return_value = {"max_tokens": 128000}
+        model_info, model = responses_api_bridge_check(
+            model="gpt-5.4",
+            custom_llm_provider="openai",
+            tools=[{"type": "function", "function": {"name": "get_capital"}}],
+            reasoning_effort="xhigh",
+        )
+
+    assert model == "gpt-5.4"
+    assert model_info.get("mode") == "responses"
+
+
+def test_responses_api_bridge_check_gpt_5_5_tools_plus_reasoning_routes_to_responses():
+    """gpt-5.5+ with both tools and reasoning_effort should route to Responses API."""
+    from litellm.main import responses_api_bridge_check
+
+    with patch("litellm.main._get_model_info_helper") as mock_get_model_info:
+        mock_get_model_info.return_value = {"max_tokens": 128000}
+        model_info, model = responses_api_bridge_check(
+            model="gpt-5.5-pro",
+            custom_llm_provider="openai",
+            tools=[{"type": "function", "function": {"name": "get_capital"}}],
+            reasoning_effort="xhigh",
+        )
+
+    assert model == "gpt-5.5-pro"
+    assert model_info.get("mode") == "responses"
+
+
+def test_responses_api_bridge_check_gpt_5_4_tools_without_reasoning_stays_chat():
+    """gpt-5.4 with tools only should not be force-routed to Responses API."""
+    from litellm.main import responses_api_bridge_check
+
+    with patch("litellm.main._get_model_info_helper") as mock_get_model_info:
+        mock_get_model_info.return_value = {"max_tokens": 128000}
+        model_info, model = responses_api_bridge_check(
+            model="gpt-5.4",
+            custom_llm_provider="openai",
+            tools=[{"type": "function", "function": {"name": "get_capital"}}],
+            reasoning_effort=None,
+        )
+
+    assert model == "gpt-5.4"
+    assert model_info.get("mode") != "responses"
+
+
 def test_responses_api_bridge_check_handles_exception():
     """Test that responses_api_bridge_check handles exceptions and still processes responses/ models."""
     from litellm.main import responses_api_bridge_check