diff --git a/tests/test_litellm/litellm_core_utils/test_streaming_response_usage_serialization.py b/tests/test_litellm/litellm_core_utils/test_streaming_response_usage_serialization.py
new file mode 100644
index 00000000000..b52b6f6841e
--- /dev/null
+++ b/tests/test_litellm/litellm_core_utils/test_streaming_response_usage_serialization.py
@@ -0,0 +1,116 @@
+"""
+Tests for pydantic-clean serialization of ResponseAPIUsage in streaming logging.
+
+When _get_assembled_streaming_response processes a ResponseCompletedEvent, it
+transforms usage from ResponseAPIUsage → Chat Completion format → back to
+ResponseAPIUsage.  The result must be a proper ResponseAPIUsage instance so that
+model_dump() on ResponsesAPIResponse doesn't emit PydanticSerializationUnexpectedValue.
+"""
+
+import warnings
+
+from litellm.litellm_core_utils.litellm_logging import Logging
+from litellm.types.llms.openai import (
+    ResponseAPIUsage,
+    ResponseCompletedEvent,
+    ResponsesAPIResponse,
+    ResponsesAPIStreamEvents,
+)
+
+
+def _make_response_completed_event(
+    input_tokens: int = 10,
+    output_tokens: int = 20,
+    total_tokens: int = 30,
+) -> ResponseCompletedEvent:
+    usage = ResponseAPIUsage(
+        input_tokens=input_tokens,
+        output_tokens=output_tokens,
+        total_tokens=total_tokens,
+    )
+    response = ResponsesAPIResponse(
+        id="resp_test",
+        created_at=1700000000,
+        output=[],
+        usage=usage,
+        status="completed",
+    )
+    return ResponseCompletedEvent(
+        type=ResponsesAPIStreamEvents.RESPONSE_COMPLETED,
+        response=response,
+    )
+
+
+def test_assembled_streaming_response_usage_is_response_api_usage():
+    """After _get_assembled_streaming_response, the usage field should be
+    a ResponseAPIUsage instance (not a raw dict)."""
+    event = _make_response_completed_event(input_tokens=5, output_tokens=15, total_tokens=20)
+
+    logging_obj = Logging.__new__(Logging)
+    result = logging_obj._get_assembled_streaming_response(
+        result=event,
+        start_time=None,
+        end_time=None,
+        is_async=False,
+        streaming_chunks=[],
+    )
+
+    assert result is not None
+    assert isinstance(result, ResponsesAPIResponse)
+    assert isinstance(result.usage, ResponseAPIUsage)
+
+
+def test_assembled_streaming_response_usage_model_dump_no_warnings():
+    """model_dump() on the response should not emit any pydantic
+    serialization warnings."""
+    event = _make_response_completed_event()
+
+    logging_obj = Logging.__new__(Logging)
+    result = logging_obj._get_assembled_streaming_response(
+        result=event,
+        start_time=None,
+        end_time=None,
+        is_async=False,
+        streaming_chunks=[],
+    )
+
+    assert result is not None
+    with warnings.catch_warnings(record=True) as caught:
+        warnings.simplefilter("always")
+        result.model_dump()
+
+    pydantic_warnings = [
+        w for w in caught if "PydanticSerializationUnexpectedValue" in str(w.message)
+    ]
+    assert len(pydantic_warnings) == 0, (
+        f"Got unexpected pydantic warnings: {pydantic_warnings}"
+    )
+
+
+def test_assembled_streaming_response_preserves_token_counts():
+    """The transformed usage should correctly map token counts from the
+    Chat Completion format back to ResponseAPIUsage fields."""
+    event = _make_response_completed_event(input_tokens=10, output_tokens=20, total_tokens=30)
+
+    logging_obj = Logging.__new__(Logging)
+    result = logging_obj._get_assembled_streaming_response(
+        result=event,
+        start_time=None,
+        end_time=None,
+        is_async=False,
+        streaming_chunks=[],
+    )
+
+    assert result is not None
+    usage = result.usage
+    # Core ResponseAPIUsage fields should be populated with the correct values
+    assert usage.input_tokens == 10
+    assert usage.output_tokens == 20
+    assert usage.total_tokens == 30
+
+    # Extra fields from the Chat Completion transform (like prompt_tokens_details)
+    # should be carried through as extras without causing serialization issues
+    usage_dict = usage.model_dump()
+    assert usage_dict["input_tokens"] == 10
+    assert usage_dict["output_tokens"] == 20
+    assert usage_dict["total_tokens"] == 30