diff --git a/litellm/llms/openai/responses/transformation.py b/litellm/llms/openai/responses/transformation.py
index 189cc7fe956..926de3e8854 100644
--- a/litellm/llms/openai/responses/transformation.py
+++ b/litellm/llms/openai/responses/transformation.py
@@ -149,7 +149,17 @@ class OpenAIResponsesAPIConfig(BaseResponsesAPIConfig):
             )
         except Exception:
             return True
-        return info["key"] in _bundled_openai_reasoning_models() or info.get("supports_reasoning") is True
+        declared: Final = info.get("supports_reasoning")
+        if declared is not None:
+            return declared
+        return info["key"] in _bundled_openai_reasoning_models()
+
+    @staticmethod
+    def _requests_reasoning_effort(reasoning: object) -> bool:
+        effort: Final = (
+            reasoning.get("effort") if isinstance(reasoning, Mapping) else getattr(reasoning, "effort", None)
+        )
+        return effort is not None
 
     @staticmethod
     def _enforce_min_max_output_tokens(max_output_tokens: "int | None") -> "int | None":
@@ -202,7 +212,7 @@ class OpenAIResponsesAPIConfig(BaseResponsesAPIConfig):
 
         if (
             self.custom_llm_provider == LlmProviders.OPENAI
-            and params.get("reasoning") is not None
+            and self._requests_reasoning_effort(params.get("reasoning"))
             and not self._supports_reasoning_param(model=model)
         ):
             if drop_params or litellm.drop_params:
@@ -210,7 +220,7 @@ class OpenAIResponsesAPIConfig(BaseResponsesAPIConfig):
             else:
                 raise litellm.UnsupportedParamsError(
                     message=(
-                        f"{model} doesn't support the `reasoning` parameter "
+                        f"{model} doesn't support `reasoning.effort` "
                         "(its model cost map entry lacks `supports_reasoning`). "
                         "To drop unsupported params set `litellm.drop_params = True`"
                     ),
diff --git a/tests/test_litellm/llms/openai/responses/test_openai_responses_transformation.py b/tests/test_litellm/llms/openai/responses/test_openai_responses_transformation.py
index 66271de7d04..c5902b32a06 100644
--- a/tests/test_litellm/llms/openai/responses/test_openai_responses_transformation.py
+++ b/tests/test_litellm/llms/openai/responses/test_openai_responses_transformation.py
@@ -2086,12 +2086,42 @@ class TestReasoningFollowsModelSupport:
                 drop_params=False,
             )
         assert excinfo.value.status_code == 400
+        assert "reasoning.effort" in str(excinfo.value)
         assert "cost map" in str(excinfo.value)
 
-    def test_azure_deployments_keep_reasoning(self, local_model_cost_map):
+    @pytest.mark.parametrize("drop_params", [True, False])
+    @pytest.mark.parametrize(
+        "reasoning",
+        [{"summary": "auto"}, {"effort": None, "summary": "auto"}, {}],
+    )
+    def test_reasoning_without_an_effort_passes_through_on_non_reasoning_models(
+        self, local_model_cost_map, monkeypatch, drop_params, reasoning
+    ):
+        monkeypatch.setattr(litellm, "drop_params", drop_params)
+        mapped = OpenAIResponsesAPIConfig().map_openai_params(
+            response_api_optional_params={"reasoning": dict(reasoning)},
+            model="gpt-4o",
+            drop_params=drop_params,
+        )
+        assert mapped["reasoning"] == reasoning
+
+    def test_an_explicit_supports_reasoning_false_beats_the_bundled_floor(self, local_model_cost_map, monkeypatch):
+        overridden = {
+            name: ({**entry, "supports_reasoning": False} if name == "o3" else entry)
+            for name, entry in litellm.model_cost.items()
+        }
+        monkeypatch.setattr(litellm, "model_cost", overridden)
+        mapped = OpenAIResponsesAPIConfig().map_openai_params(
+            response_api_optional_params={"reasoning": {"effort": "medium"}},
+            model="o3",
+            drop_params=True,
+        )
+        assert "reasoning" not in mapped
+
+    def test_azure_deployments_keep_reasoning_even_on_a_non_reasoning_model_name(self, local_model_cost_map):
         mapped = AzureOpenAIResponsesAPIConfig().map_openai_params(
             response_api_optional_params={"reasoning": {"effort": "medium"}},
-            model="my-o3-deployment",
+            model="gpt-4o",
             drop_params=True,
         )
         assert mapped["reasoning"] == {"effort": "medium"}