fix(responses): pass extra_body through responses-to-completion bridge (#20982)

extra_body was silently dropped when calling litellm.responses() for
models using the completion bridge. The parameter is accepted as an
explicit kwarg in responses(), so it's not included in **kwargs, and the
handler/transformation functions didn't accept or forward it.

Changes:
- Pass extra_body from responses() to response_api_handler()
- Add extra_body parameter to response_api_handler() and forward it to
  the transformation function
- Add extra_body parameter to
  transform_responses_api_request_to_chat_completion_request() and
  include it in the litellm_completion_request dict
- Add 5 regression tests verifying extra_body flows correctly

litellm/responses/litellm_completion_transformation/handler.py

@@ -31,6 +31,7 @@ class LiteLLMCompletionTransformationHandler:
         _is_async: bool = False,
         stream: Optional[bool] = None,
         extra_headers: Optional[Dict[str, Any]] = None,
+        extra_body: Optional[Dict[str, Any]] = None,
         **kwargs,
     ) -> Union[
         ResponsesAPIResponse,
@@ -47,6 +48,7 @@ class LiteLLMCompletionTransformationHandler:
                 custom_llm_provider=custom_llm_provider,
                 stream=stream,
                 extra_headers=extra_headers,
+                extra_body=extra_body,
                 **kwargs,
             )
         )

litellm/responses/litellm_completion_transformation/transformation.py

@@ -161,6 +161,7 @@ class LiteLLMCompletionResponsesConfig:
         custom_llm_provider: Optional[str] = None,
         stream: Optional[bool] = None,
         extra_headers: Optional[Dict[str, Any]] = None,
+        extra_body: Optional[Dict[str, Any]] = None,
         **kwargs,
     ) -> dict:
         """
@@ -215,6 +216,7 @@ class LiteLLMCompletionResponsesConfig:
             # litellm specific params
             "custom_llm_provider": custom_llm_provider,
             "extra_headers": extra_headers,
+            "extra_body": extra_body,
         }
 
         # Responses API `Completed` events require usage, we pass `stream_options` to litellm.completion to include usage

litellm/responses/main.py

@@ -681,6 +681,7 @@ def responses(
                 _is_async=_is_async,
                 stream=stream,
                 extra_headers=extra_headers,
+                extra_body=extra_body,
                 **kwargs,
             )
 

tests/test_litellm/test_extra_body_responses_bridge.py

@@ -0,0 +1,100 @@
+"""
+Regression tests for #20982 — ``extra_body`` is not passed through the
+responses-to-completion bridge.
+
+When calling ``litellm.responses()`` with ``extra_body`` for a model
+that uses the completion bridge, the ``extra_body`` parameter was silently
+dropped because:
+1. It's an explicit parameter, so it's not in ``**kwargs``
+2. The handler and transformation functions didn't accept/forward it
+"""
+
+import os
+import sys
+
+sys.path.insert(
+    0, os.path.abspath("../..")
+)  # Adds the parent directory to the system path
+
+from litellm.responses.litellm_completion_transformation.transformation import (
+    LiteLLMCompletionResponsesConfig,
+)
+
+
+class TestExtraBodyResponsesBridge:
+    """Verify ``extra_body`` flows through the responses-to-completion
+    transformation."""
+
+    def test_extra_body_included_in_completion_request(self):
+        """extra_body should appear in the transformed completion request."""
+        extra_body = {"provider": {"order": ["Together"], "allow_fallbacks": False}}
+        result = LiteLLMCompletionResponsesConfig.transform_responses_api_request_to_chat_completion_request(
+            model="openrouter/test-model",
+            input="Hello",
+            responses_api_request={},
+            custom_llm_provider="openrouter",
+            stream=False,
+            extra_headers=None,
+            extra_body=extra_body,
+        )
+        assert result.get("extra_body") == extra_body
+
+    def test_extra_body_none_by_default(self):
+        """When not provided, extra_body should be None in the request."""
+        result = LiteLLMCompletionResponsesConfig.transform_responses_api_request_to_chat_completion_request(
+            model="openrouter/test-model",
+            input="Hello",
+            responses_api_request={},
+            custom_llm_provider="openrouter",
+            stream=False,
+            extra_headers=None,
+        )
+        assert result.get("extra_body") is None
+
+    def test_extra_body_preserved_with_other_params(self):
+        """extra_body should coexist with other parameters like
+        extra_headers, temperature, etc."""
+        extra_body = {"custom_key": "custom_value"}
+        extra_headers = {"X-Custom": "header"}
+        result = LiteLLMCompletionResponsesConfig.transform_responses_api_request_to_chat_completion_request(
+            model="test-model",
+            input="Hello",
+            responses_api_request={"temperature": 0.7, "max_output_tokens": 100},
+            custom_llm_provider=None,
+            stream=True,
+            extra_headers=extra_headers,
+            extra_body=extra_body,
+        )
+        assert result["extra_body"] == extra_body
+        assert result["extra_headers"] == extra_headers
+        assert result["temperature"] == 0.7
+        assert result["max_tokens"] == 100
+
+    def test_extra_body_empty_dict(self):
+        """An empty extra_body dict should still be passed through."""
+        result = LiteLLMCompletionResponsesConfig.transform_responses_api_request_to_chat_completion_request(
+            model="test-model",
+            input="Hello",
+            responses_api_request={},
+            extra_body={},
+        )
+        assert result.get("extra_body") == {}
+
+    def test_extra_body_nested_dict(self):
+        """Deeply nested extra_body should be preserved as-is."""
+        extra_body = {
+            "provider": {
+                "order": ["Together", "Fireworks"],
+                "allow_fallbacks": False,
+                "quantizations": ["fp16"],
+            },
+            "transforms": ["middle-out"],
+        }
+        result = LiteLLMCompletionResponsesConfig.transform_responses_api_request_to_chat_completion_request(
+            model="openrouter/model",
+            input="Test",
+            responses_api_request={},
+            extra_body=extra_body,
+        )
+        assert result["extra_body"] == extra_body
+        assert result["extra_body"]["provider"]["order"] == ["Together", "Fireworks"]