From 31f270cd6c533f87920e9503a92ad87db8b444cc Mon Sep 17 00:00:00 2001
From: Mohammad Javad Naderi <mjnaderi@gmail.com>
Date: Thu, 23 Jul 2026 05:06:21 +0330
Subject: [PATCH] fix(logging): preserve Responses API streaming usage

---
 litellm/litellm_core_utils/litellm_logging.py | 15 -------
 .../test_litellm_logging.py                   | 44 +++++++++++++++++++
 2 files changed, 44 insertions(+), 15 deletions(-)

diff --git a/litellm/litellm_core_utils/litellm_logging.py b/litellm/litellm_core_utils/litellm_logging.py
index a72d46e3fe8..b6240588298 100644
--- a/litellm/litellm_core_utils/litellm_logging.py
+++ b/litellm/litellm_core_utils/litellm_logging.py
@@ -3409,21 +3409,6 @@ class Logging(LiteLLMLoggingBaseClass):
             result,
             (ResponseCompletedEvent, ResponseIncompleteEvent, ResponseFailedEvent),
         ):
-            ## return unified Usage object
-            if isinstance(result.response.usage, ResponseAPIUsage):
-                transformed_usage: Final = ResponseAPILoggingUtils._transform_response_api_usage_to_chat_usage(
-                    result.response.usage
-                )
-                # Set as dict instead of Usage object so model_dump() serializes it correctly
-                setattr(
-                    result.response,
-                    "usage",
-                    (
-                        transformed_usage.model_dump()
-                        if hasattr(transformed_usage, "model_dump")
-                        else dict(transformed_usage)
-                    ),
-                )
             return result.response
         else:
             return None
diff --git a/tests/test_litellm/litellm_core_utils/test_litellm_logging.py b/tests/test_litellm/litellm_core_utils/test_litellm_logging.py
index 28a6c8dd18d..7004c1a7846 100644
--- a/tests/test_litellm/litellm_core_utils/test_litellm_logging.py
+++ b/tests/test_litellm/litellm_core_utils/test_litellm_logging.py
@@ -2721,6 +2721,50 @@ def test_get_assembled_streaming_response_returns_result_for_streaming():
     assert assembled is result
 
 
+def test_get_assembled_streaming_response_preserves_responses_usage():
+    import datetime
+
+    from litellm.litellm_core_utils.litellm_logging import StandardLoggingPayloadSetup
+    from litellm.types.llms.openai import (
+        ResponseAPIUsage,
+        ResponseCompletedEvent,
+        ResponsesAPIResponse,
+        ResponsesAPIStreamEvents,
+    )
+
+    logging_obj = _make_logging_obj(stream=True)
+    response = ResponsesAPIResponse(
+        id="resp-1",
+        created_at=1700000000,
+        output=[],
+        usage=ResponseAPIUsage(
+            input_tokens=4,
+            output_tokens=5,
+            total_tokens=9,
+        ),
+    )
+    event = ResponseCompletedEvent(
+        type=ResponsesAPIStreamEvents.RESPONSE_COMPLETED,
+        response=response,
+    )
+
+    assembled = logging_obj._get_assembled_streaming_response(
+        result=event,
+        start_time=datetime.datetime.now(),
+        end_time=datetime.datetime.now(),
+        is_async=True,
+        streaming_chunks=[],
+    )
+
+    assert isinstance(assembled, ResponsesAPIResponse)
+    assembled.model_dump(warnings="error")
+    assert isinstance(assembled.usage, ResponseAPIUsage)
+    logging_usage = StandardLoggingPayloadSetup.get_usage_from_response_obj(assembled.model_dump())
+    assert logging_usage.prompt_tokens == 4
+    assert logging_usage.completion_tokens == 5
+    assert logging_usage.total_tokens == 9
+
+
 def test_streaming_success_handler_includes_vertex_ai_metadata_in_standard_logging():
     """Assembled streaming responses should include Vertex AI metadata in logging payload."""
     import datetime