Merge abc79dd430 into d7620c77a1

litellm/llms/openai/responses/transformation.py

@@ -484,8 +484,8 @@ class OpenAIResponsesAPIConfig(BaseResponsesAPIConfig):
         )
         if sanitized_tools is not None:
             response_api_optional_request_params["tools"] = sanitized_tools
-        final_request_params: Final = dict(
-            ResponsesAPIRequestParams(model=model, input=replay_safe_input, **response_api_optional_request_params)
+        final_request_params: Final = dict[str, object](
+            model=model, input=replay_safe_input, **response_api_optional_request_params
         )
 
         return final_request_params
@@ -1197,9 +1197,7 @@ class OpenAIResponsesAPIConfig(BaseResponsesAPIConfig):
         )
         if sanitized_tools is not None:
             response_api_optional_request_params["tools"] = sanitized_tools
-        data: Final = dict(
-            ResponsesAPIRequestParams(model=model, input=replay_safe_input, **response_api_optional_request_params)
-        )
+        data: Final = dict[str, object](model=model, input=replay_safe_input, **response_api_optional_request_params)
 
         return url, data
 

litellm/types/llms/openai.py

@@ -1323,6 +1323,26 @@ class ResponsesAPIStreamOptions(TypedDict, total=False):
     include_obfuscation: bool
 
 
+class ResponsesAPIModerationPolicyEntry(TypedDict, total=False):
+    """Moderation policy for either the input or the output of a response."""
+
+    mode: Required[ReadOnly[Literal["score", "block"]]]
+
+
+class ResponsesAPIModerationPolicy(TypedDict, total=False):
+    """Policy applied to moderated response input and output."""
+
+    input: ReadOnly[ResponsesAPIModerationPolicyEntry | None]
+    output: ReadOnly[ResponsesAPIModerationPolicyEntry | None]
+
+
+class ResponsesAPIModeration(TypedDict, total=False):
+    """Configuration for running moderation on the input and output of a response."""
+
+    model: Required[ReadOnly[str]]
+    policy: ReadOnly[ResponsesAPIModerationPolicy | None]
+
+
 class ResponsesAPIOptionalRequestParams(TypedDict, total=False):
     """TypedDict for Optional parameters supported by the responses API."""
 
@@ -1355,6 +1375,8 @@ class ResponsesAPIOptionalRequestParams(TypedDict, total=False):
     partial_images: int | None  # Number of partial images to generate (1-3) for streaming image generation
     context_management: list[ContextManagementEntry] | None
     """Context management configuration. E.g. [{\"type\": \"compaction\", \"compact_threshold\": 200000}] for server-side compaction (minimum 1000)."""
+    moderation: ReadOnly[ResponsesAPIModeration | None]
+    """Moderation configuration for the response input and output. E.g. {\"model\": \"omni-moderation-latest\"}."""
 
 
 class ResponsesAPIRequestParams(ResponsesAPIOptionalRequestParams, total=False):

tests/unit/responses/test_responses_api_request_body.py

@@ -228,6 +228,56 @@ async def test_aresponses_drops_stream_options():
         assert "stream_options" not in request_body
 
 
+@pytest.mark.asyncio
+async def test_aresponses_forwards_moderation():
+    """OpenAI's top-level `moderation` param must reach the wire instead of being silently dropped."""
+    moderation: Final = {
+        "model": "omni-moderation-latest",
+        "policy": {"input": {"mode": "score"}, "output": {"mode": "block"}},
+    }
+    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_moderation_test", "gpt-5.4"), 200)
+
+        await litellm.aresponses(
+            model="openai/gpt-5.4",
+            api_key="fake-api-key",
+            input="hi",
+            moderation=moderation,
+        )
+
+        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["moderation"] == moderation
+
+
+@pytest.mark.asyncio
+async def test_aresponses_omits_moderation_when_none():
+    """An explicit moderation=None is a default value and must not be sent to the provider."""
+    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_moderation_none_test", "gpt-5.4"), 200
+        )
+
+        await litellm.aresponses(
+            model="openai/gpt-5.4",
+            api_key="fake-api-key",
+            input="hi",
+            moderation=None,
+        )
+
+        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 "moderation" not in request_body
+
+
 @pytest.mark.asyncio
 async def test_aresponses_forwards_non_enum_reasoning_effort(
     monkeypatch: pytest.MonkeyPatch, respx_mock: respx.MockRouter

ui/litellm-dashboard/src/lib/http/schema.d.ts

@@ -43893,6 +43893,34 @@ export interface components {
              */
             type: "tool_search_output";
         };
+        /**
+         * ResponsesAPIModeration
+         * @description Configuration for running moderation on the input and output of a response.
+         */
+        ResponsesAPIModeration: {
+            /** Model */
+            model: string;
+            policy?: components["schemas"]["ResponsesAPIModerationPolicy"] | null;
+        };
+        /**
+         * ResponsesAPIModerationPolicy
+         * @description Policy applied to moderated response input and output.
+         */
+        ResponsesAPIModerationPolicy: {
+            input?: components["schemas"]["ResponsesAPIModerationPolicyEntry"] | null;
+            output?: components["schemas"]["ResponsesAPIModerationPolicyEntry"] | null;
+        };
+        /**
+         * ResponsesAPIModerationPolicyEntry
+         * @description Moderation policy for either the input or the output of a response.
+         */
+        ResponsesAPIModerationPolicyEntry: {
+            /**
+             * Mode
+             * @enum {string}
+             */
+            mode: "score" | "block";
+        };
         /**
          * ResponsesAPIRequestParams
          * @description TypedDict for request parameters supported by the responses API.
@@ -43918,6 +43946,7 @@ export interface components {
             } | null;
             /** Model */
             model: string;
+            moderation?: components["schemas"]["ResponsesAPIModeration"] | null;
             /** Parallel Tool Calls */
             parallel_tool_calls?: boolean | null;
             /** Partial Images */
@@ -66337,6 +66366,7 @@ export interface operations {
                     } | null;
                     /** Model */
                     model: string;
+                    moderation?: components["schemas"]["ResponsesAPIModeration"] | null;
                     /** Parallel Tool Calls */
                     parallel_tool_calls?: boolean | null;
                     /** Partial Images */
@@ -69801,6 +69831,7 @@ export interface operations {
                     } | null;
                     /** Model */
                     model: string;
+                    moderation?: components["schemas"]["ResponsesAPIModeration"] | null;
                     /** Parallel Tool Calls */
                     parallel_tool_calls?: boolean | null;
                     /** Partial Images */
@@ -80458,6 +80489,7 @@ export interface operations {
                     } | null;
                     /** Model */
                     model: string;
+                    moderation?: components["schemas"]["ResponsesAPIModeration"] | null;
                     /** Parallel Tool Calls */
                     parallel_tool_calls?: boolean | null;
                     /** Partial Images */