From dc30e1816d89c7124db6f20bdb6d68296b5ab6d2 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Tue, 11 Aug 2026 20:11:53 -0700
Subject: [PATCH] refactor(passthrough): move Responses stream terminal-event
 parsing into OpenAI provider config

Addresses review feedback: the ResponseCompletedEvent SSE parsing now lives
in OpenAIResponsesAPIConfig next to the other Responses stream event handling,
and the proxy logging handler calls it. Adds coverage for streams that end
without a response.completed event.
---
 .../llms/openai/responses/transformation.py   |  9 +++++++
 .../openai_passthrough_logging_handler.py     | 13 ++--------
 ...test_openai_passthrough_logging_handler.py | 24 +++++++++++++++++++
 3 files changed, 35 insertions(+), 11 deletions(-)

diff --git a/litellm/llms/openai/responses/transformation.py b/litellm/llms/openai/responses/transformation.py
index f12a034b6ad..11a82172316 100644
--- a/litellm/llms/openai/responses/transformation.py
+++ b/litellm/llms/openai/responses/transformation.py
@@ -353,6 +353,15 @@ class OpenAIResponsesAPIConfig(BaseResponsesAPIConfig):
             )
             return event_pydantic_model.model_construct(**parsed_chunk)
 
+    @staticmethod
+    def parse_completed_response_from_stream_chunks(all_chunks: list[str]) -> ResponsesAPIResponse | None:
+        for chunk_str in reversed(all_chunks):
+            try:
+                return ResponseCompletedEvent.model_validate_json(chunk_str.removeprefix("data: ")).response
+            except ValueError:
+                continue
+        return None
+
     @staticmethod
     def get_event_model_class(event_type: str) -> Any:
         """
diff --git a/litellm/proxy/pass_through_endpoints/llm_provider_handlers/openai_passthrough_logging_handler.py b/litellm/proxy/pass_through_endpoints/llm_provider_handlers/openai_passthrough_logging_handler.py
index 28aa09d5a94..97c2c682a58 100644
--- a/litellm/proxy/pass_through_endpoints/llm_provider_handlers/openai_passthrough_logging_handler.py
+++ b/litellm/proxy/pass_through_endpoints/llm_provider_handlers/openai_passthrough_logging_handler.py
@@ -26,7 +26,7 @@ from litellm.proxy.pass_through_endpoints.llm_provider_handlers.base_passthrough
 from litellm.proxy.pass_through_endpoints.success_handler import (
     PassThroughEndpointLogging,
 )
-from litellm.types.llms.openai import ResponseCompletedEvent, ResponsesAPIResponse
+from litellm.types.llms.openai import ResponsesAPIResponse
 from litellm.types.passthrough_endpoints.pass_through_endpoints import (
     EndpointType,
     PassthroughStandardLoggingPayload,
@@ -518,15 +518,6 @@ class OpenAIPassthroughLoggingHandler(BasePassthroughLoggingHandler):
             verbose_proxy_logger.error("Error building complete streaming response: %s", e)
             return None
 
-    @staticmethod
-    def _build_complete_streaming_responses_response(all_chunks: list[str]) -> ResponsesAPIResponse | None:
-        for chunk_str in reversed(all_chunks):
-            try:
-                return ResponseCompletedEvent.model_validate_json(chunk_str.removeprefix("data: ")).response
-            except ValueError:
-                continue
-        return None
-
     @staticmethod
     def _handle_logging_openai_collected_chunks(
         litellm_logging_obj: LiteLLMLoggingObj,
@@ -551,7 +542,7 @@ class OpenAIPassthroughLoggingHandler(BasePassthroughLoggingHandler):
             handler: Final = OpenAIPassthroughLoggingHandler()
             handler_instance: Final = handler
             complete_response: Final = (
-                handler._build_complete_streaming_responses_response(all_chunks=all_chunks)
+                OpenAIResponsesAPIConfig.parse_completed_response_from_stream_chunks(all_chunks=all_chunks)
                 if is_responses
                 else handler._build_complete_streaming_response(
                     all_chunks=all_chunks,
diff --git a/tests/test_litellm/proxy/pass_through_endpoints/llm_provider_handlers/test_openai_passthrough_logging_handler.py b/tests/test_litellm/proxy/pass_through_endpoints/llm_provider_handlers/test_openai_passthrough_logging_handler.py
index 9e5da39af88..6b48d575281 100644
--- a/tests/test_litellm/proxy/pass_through_endpoints/llm_provider_handlers/test_openai_passthrough_logging_handler.py
+++ b/tests/test_litellm/proxy/pass_through_endpoints/llm_provider_handlers/test_openai_passthrough_logging_handler.py
@@ -593,6 +593,30 @@ class TestOpenAIPassthroughLoggingHandler:
             call_type="responses",
         )
 
+    @patch(f"{OpenAIPassthroughLoggingHandler.__module__}.get_standard_logging_object_payload")
+    @patch("litellm.completion_cost")
+    def test_streaming_responses_without_completed_event_returns_none(
+        self, mock_completion_cost, mock_get_standard_logging
+    ):
+        logging_obj = self._create_mock_logging_obj()
+
+        result = OpenAIPassthroughLoggingHandler._handle_logging_openai_collected_chunks(
+            litellm_logging_obj=logging_obj,
+            passthrough_success_handler_obj=MagicMock(),
+            url_route="https://api.openai.com/v1/responses",
+            request_body={"model": "gpt-4o-mini", "stream": True},
+            endpoint_type=MagicMock(),
+            start_time=self.start_time,
+            all_chunks=[
+                'data: {"type": "response.created", "sequence_number": 0}',
+                'data: {"type": "response.output_text.delta", "sequence_number": 1, "delta": "OK"}',
+            ],
+            end_time=self.end_time,
+        )
+
+        assert result == {"result": None, "kwargs": {}}
+        mock_completion_cost.assert_not_called()
+
     @patch("litellm.completion_cost")
     @patch(
         "litellm.litellm_core_utils.litellm_logging.get_standard_logging_object_payload"