From a5be969ae77e77de07d4ec9caea409fac8ccf48e Mon Sep 17 00:00:00 2001
From: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Date: Tue, 28 Jul 2026 03:09:15 +0000
Subject: [PATCH 1/2] test(logging): lock in streaming response_cost survival
 across sync success handler

---
 .../test_litellm_logging.py                   | 33 +++++++++++++++++++
 1 file changed, 33 insertions(+)

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 edc257f4c3f..cf888974928 100644
--- a/tests/test_litellm/litellm_core_utils/test_litellm_logging.py
+++ b/tests/test_litellm/litellm_core_utils/test_litellm_logging.py
@@ -2796,6 +2796,39 @@ async def test_async_success_handler_preserves_response_cost_for_pass_through_en
     assert slo is not None
     assert slo["response_cost"] > 0
 
+@pytest.mark.asyncio
+async def test_sync_success_handler_preserves_streaming_cost_computed_by_async_handler():
+    """Regression for #34875: the sync success_handler used to reset response_cost to None
+    for streams, so custom callbacks reading kwargs["response_cost"] saw None whenever it ran
+    after the async handler had already computed the cost on the shared model_call_details."""
+    import litellm
+
+    litellm.callbacks = []
+    litellm.success_callback = []
+
+    response = await litellm.acompletion(
+        model="gpt-4o-mini",
+        messages=[{"role": "user", "content": "hi"}],
+        stream=True,
+        mock_response="hello there",
+        stream_options={"include_usage": True},
+    )
+    logging_obj = response.logging_obj
+    async for _chunk in response:
+        pass
+    await asyncio.sleep(0.5)
+
+    cost_from_async_handler = logging_obj.model_call_details.get("response_cost")
+    assert cost_from_async_handler is not None and cost_from_async_handler > 0
+
+    assembled = logging_obj.model_call_details.get(
+        "async_complete_streaming_response"
+    ) or logging_obj.model_call_details.get("complete_streaming_response")
+
+    for _ in range(2):
+        logging_obj.success_handler(assembled, None, None, False)
+        assert logging_obj.model_call_details["response_cost"] == cost_from_async_handler
+
 
 def test_process_hidden_params_recalculates_cost_after_failure_handler_zero():
     """

From 2a5bd9562dda19c5050d538f4287ef30555cbb42 Mon Sep 17 00:00:00 2001
From: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Date: Tue, 28 Jul 2026 15:37:13 +0000
Subject: [PATCH 2/2] test(logging): poll for the async streaming cost instead
 of a fixed sleep

---
 .../litellm_core_utils/test_litellm_logging.py      | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

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 cf888974928..67c976c6ab7 100644
--- a/tests/test_litellm/litellm_core_utils/test_litellm_logging.py
+++ b/tests/test_litellm/litellm_core_utils/test_litellm_logging.py
@@ -1,6 +1,7 @@
 import os
 import sys
 import asyncio
+from typing import Optional
 from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
@@ -2796,6 +2797,7 @@ async def test_async_success_handler_preserves_response_cost_for_pass_through_en
     assert slo is not None
     assert slo["response_cost"] > 0
 
+
 @pytest.mark.asyncio
 async def test_sync_success_handler_preserves_streaming_cost_computed_by_async_handler():
     """Regression for #34875: the sync success_handler used to reset response_cost to None
@@ -2816,9 +2818,16 @@ async def test_sync_success_handler_preserves_streaming_cost_computed_by_async_h
     logging_obj = response.logging_obj
     async for _chunk in response:
         pass
-    await asyncio.sleep(0.5)
 
-    cost_from_async_handler = logging_obj.model_call_details.get("response_cost")
+    async def wait_for_async_handler_cost() -> Optional[float]:
+        for _ in range(200):
+            cost = logging_obj.model_call_details.get("response_cost")
+            if cost:
+                return cost
+            await asyncio.sleep(0.05)
+        return None
+
+    cost_from_async_handler = await wait_for_async_handler_cost()
     assert cost_from_async_handler is not None and cost_from_async_handler > 0
 
     assembled = logging_obj.model_call_details.get(