diff --git a/litellm/responses/main.py b/litellm/responses/main.py
index 944cf58df1c..a2441adbaac 100644
--- a/litellm/responses/main.py
+++ b/litellm/responses/main.py
@@ -1028,7 +1028,10 @@ def responses(
                 local_vars["reasoning"] = _mapped
         # Get ResponsesAPIOptionalRequestParams with only valid parameters
         response_api_optional_params: ResponsesAPIOptionalRequestParams = (
-            ResponsesAPIRequestUtils.get_requested_response_api_optional_param(local_vars)
+            ResponsesAPIRequestUtils.get_requested_response_api_optional_param(
+                local_vars,
+                allowed_openai_params=allowed_openai_params,
+            )
         )
 
         _file_search_dispatch = _responses_try_dispatch_emulated_file_search(
diff --git a/litellm/responses/utils.py b/litellm/responses/utils.py
index 429ddeef36a..3a92663e2f1 100644
--- a/litellm/responses/utils.py
+++ b/litellm/responses/utils.py
@@ -7,6 +7,7 @@ from typing import (
     List,
     Mapping,
     Optional,
+    Sequence,
     Type,
     Union,
     cast,
@@ -185,19 +186,26 @@ class ResponsesAPIRequestUtils:
     @staticmethod
     def get_requested_response_api_optional_param(
         params: Dict[str, Any],
+        allowed_openai_params: Sequence[str] | None = None,
     ) -> ResponsesAPIOptionalRequestParams:
         """
-        Filter parameters to only include those defined in ResponsesAPIOptionalRequestParams.
+        Filter parameters to only include those defined in ResponsesAPIOptionalRequestParams,
+        plus any request/deployment level `allowed_openai_params` the caller opted into.
 
         Args:
             params: Dictionary of parameters to filter
+            allowed_openai_params: Extra request fields to keep even though they are not
+                part of the Responses API spec
 
         Returns:
             ResponsesAPIOptionalRequestParams instance with only the valid parameters
         """
+        from litellm.types.utils import all_litellm_params
         from litellm.utils import PreProcessNonDefaultParams
 
-        valid_keys = get_type_hints(ResponsesAPIOptionalRequestParams).keys()
+        valid_keys = frozenset(get_type_hints(ResponsesAPIOptionalRequestParams)) | frozenset(
+            param for param in allowed_openai_params or () if param not in all_litellm_params
+        )
         custom_llm_provider = params.pop("custom_llm_provider", None)
         special_params = params.pop("kwargs", {})
 
diff --git a/tests/test_litellm/responses/test_responses_api_request_body.py b/tests/test_litellm/responses/test_responses_api_request_body.py
index 83b9c34636e..040eae7cadc 100644
--- a/tests/test_litellm/responses/test_responses_api_request_body.py
+++ b/tests/test_litellm/responses/test_responses_api_request_body.py
@@ -246,6 +246,84 @@ async def test_aresponses_keeps_include_obfuscation_in_stream_options():
         assert request_body["stream_options"] == {"include_obfuscation": False}
 
 
+@pytest.mark.asyncio
+async def test_aresponses_allowed_openai_params_forwards_extension_param():
+    """
+    allowed_openai_params must preserve request fields that are not part of the
+    Responses API spec, so OpenAI-compatible upstreams that require them still
+    receive them. Regression test for https://github.com/BerriAI/litellm/issues/34926
+    """
+    with patch(
+        "litellm.llms.custom_httpx.http_handler.AsyncHTTPHandler.post",
+        new_callable=AsyncMock,
+    ) as mock_post:
+        mock_post.return_value = MockResponse(
+            _minimal_responses_api_payload("resp_allowed_params_test", "gpt-5.5"), 200
+        )
+
+        await litellm.aresponses(
+            model="openai/gpt-5.5",
+            api_key="fake-api-key",
+            input="ping",
+            allowed_openai_params=["client_metadata"],
+            client_metadata={"probe": "keep-me"},
+        )
+
+        mock_post.assert_called_once()
+        post_kwargs = mock_post.call_args.kwargs
+        request_body = post_kwargs["json"] if "json" in post_kwargs else json.loads(post_kwargs["data"])
+        assert request_body["client_metadata"] == {"probe": "keep-me"}
+
+
+@pytest.mark.asyncio
+async def test_aresponses_drops_extension_param_without_allowlist():
+    """Without an allowlist, unknown extension params stay off the wire."""
+    with patch(
+        "litellm.llms.custom_httpx.http_handler.AsyncHTTPHandler.post",
+        new_callable=AsyncMock,
+    ) as mock_post:
+        mock_post.return_value = MockResponse(
+            _minimal_responses_api_payload("resp_no_allowlist_test", "gpt-5.5"), 200
+        )
+
+        await litellm.aresponses(
+            model="openai/gpt-5.5",
+            api_key="fake-api-key",
+            input="ping",
+            client_metadata={"probe": "keep-me"},
+        )
+
+        mock_post.assert_called_once()
+        post_kwargs = mock_post.call_args.kwargs
+        request_body = post_kwargs["json"] if "json" in post_kwargs else json.loads(post_kwargs["data"])
+        assert "client_metadata" not in request_body
+
+
+@pytest.mark.asyncio
+async def test_aresponses_allowed_openai_params_cannot_leak_litellm_params():
+    """allowed_openai_params must never promote internal litellm params (e.g. credentials) into the request body."""
+    with patch(
+        "litellm.llms.custom_httpx.http_handler.AsyncHTTPHandler.post",
+        new_callable=AsyncMock,
+    ) as mock_post:
+        mock_post.return_value = MockResponse(
+            _minimal_responses_api_payload("resp_no_credential_leak_test", "gpt-5.5"), 200
+        )
+
+        await litellm.aresponses(
+            model="openai/gpt-5.5",
+            api_key="fake-api-key",
+            input="ping",
+            allowed_openai_params=["api_key", "api_base"],
+        )
+
+        mock_post.assert_called_once()
+        post_kwargs = mock_post.call_args.kwargs
+        request_body = post_kwargs["json"] if "json" in post_kwargs else json.loads(post_kwargs["data"])
+        assert "api_key" not in request_body
+        assert "api_base" not in request_body
+
+
 @pytest.mark.asyncio
 async def test_aresponses_request_level_drop_params_drops_bedrock_mantle_service_tier(
     monkeypatch,