diff --git a/src/agents/models/openai_responses.py b/src/agents/models/openai_responses.py
index ce7a08b7f..27760e9e2 100644
--- a/src/agents/models/openai_responses.py
+++ b/src/agents/models/openai_responses.py
@@ -271,9 +271,15 @@ async def _fetch_response(
         should_omit_model = prompt is not None and not self._model_is_explicit
         model_param: str | ChatModel | Omit = self.model if not should_omit_model else omit
         should_omit_tools = prompt is not None and len(converted_tools_payload) == 0
+        # In prompt-managed tool flows without local tools payload, omit only named tool choices
+        # that must match an explicit tool list. Keep control literals like "none"/"required".
+        should_omit_tool_choice = should_omit_tools and isinstance(tool_choice, dict)
         tools_param: list[ToolParam] | Omit = (
             converted_tools_payload if not should_omit_tools else omit
         )
+        tool_choice_param: response_create_params.ToolChoice | Omit = (
+            tool_choice if not should_omit_tool_choice else omit
+        )
 
         include_set: set[str] = set(converted_tools.includes)
         if model_settings.response_include is not None:
@@ -300,7 +306,7 @@ async def _fetch_response(
                 f"{input_json}\n"
                 f"Tools:\n{tools_json}\n"
                 f"Stream: {stream}\n"
-                f"Tool choice: {tool_choice}\n"
+                f"Tool choice: {tool_choice_param}\n"
                 f"Response format: {response_format}\n"
                 f"Previous response id: {previous_response_id}\n"
                 f"Conversation id: {conversation_id}\n"
@@ -330,7 +336,7 @@ async def _fetch_response(
             top_p=self._non_null_or_omit(model_settings.top_p),
             truncation=self._non_null_or_omit(model_settings.truncation),
             max_output_tokens=self._non_null_or_omit(model_settings.max_tokens),
-            tool_choice=tool_choice,
+            tool_choice=tool_choice_param,
             parallel_tool_calls=parallel_tool_calls,
             stream=cast(Any, stream_param),
             extra_headers=self._merge_headers(model_settings),
diff --git a/tests/test_openai_responses.py b/tests/test_openai_responses.py
index ecd509ac6..b896ea819 100644
--- a/tests/test_openai_responses.py
+++ b/tests/test_openai_responses.py
@@ -135,3 +135,76 @@ def __init__(self):
     )
 
     assert called_kwargs["tools"] is omit
+
+
+@pytest.mark.allow_call_model_methods
+@pytest.mark.asyncio
+async def test_prompt_id_omits_tool_choice_when_no_tools_configured():
+    called_kwargs: dict[str, Any] = {}
+
+    class DummyResponses:
+        async def create(self, **kwargs):
+            nonlocal called_kwargs
+            called_kwargs = kwargs
+            return get_response_obj([])
+
+    class DummyResponsesClient:
+        def __init__(self):
+            self.responses = DummyResponses()
+
+    model = OpenAIResponsesModel(
+        model="gpt-4",
+        openai_client=DummyResponsesClient(),  # type: ignore[arg-type]
+        model_is_explicit=False,
+    )
+
+    await model.get_response(
+        system_instructions=None,
+        input="hi",
+        model_settings=ModelSettings(tool_choice="web_search_preview"),
+        tools=[],
+        output_schema=None,
+        handoffs=[],
+        tracing=ModelTracing.DISABLED,
+        prompt={"id": "pmpt_123"},
+    )
+
+    assert called_kwargs["tools"] is omit
+    assert called_kwargs["tool_choice"] is omit
+
+
+@pytest.mark.allow_call_model_methods
+@pytest.mark.asyncio
+@pytest.mark.parametrize("tool_choice", ["none", "required"])
+async def test_prompt_id_keeps_literal_tool_choice_without_local_tools(tool_choice: str):
+    called_kwargs: dict[str, Any] = {}
+
+    class DummyResponses:
+        async def create(self, **kwargs):
+            nonlocal called_kwargs
+            called_kwargs = kwargs
+            return get_response_obj([])
+
+    class DummyResponsesClient:
+        def __init__(self):
+            self.responses = DummyResponses()
+
+    model = OpenAIResponsesModel(
+        model="gpt-4",
+        openai_client=DummyResponsesClient(),  # type: ignore[arg-type]
+        model_is_explicit=False,
+    )
+
+    await model.get_response(
+        system_instructions=None,
+        input="hi",
+        model_settings=ModelSettings(tool_choice=tool_choice),
+        tools=[],
+        output_schema=None,
+        handoffs=[],
+        tracing=ModelTracing.DISABLED,
+        prompt={"id": "pmpt_123"},
+    )
+
+    assert called_kwargs["tools"] is omit
+    assert called_kwargs["tool_choice"] == tool_choice