Merge pull request #42385 from BerriAI/litellm_fix_responses_stream_cost_breakdown

fix(logging): price terminal Responses stream events from their inner response

litellm/litellm_core_utils/litellm_logging.py

@@ -1757,13 +1757,21 @@ class Logging(LiteLLMLoggingBaseClass):
         if transformed_result is not None:
             result = transformed_result
 
-        result_hidden_params: Final = getattr(result, "_hidden_params", None) or MappingProxyType({})
+        priced_result: Final = (
+            result.response
+            if isinstance(result, (ResponseCompletedEvent, ResponseIncompleteEvent, ResponseFailedEvent))
+            else result
+        )
+
+        result_hidden_params: Final = getattr(priced_result, "_hidden_params", None) or MappingProxyType({})
         result_additional_headers: Final = (
             result_hidden_params.get("additional_headers")
             if isinstance(result_hidden_params, dict)
             else getattr(result_hidden_params, "additional_headers", None)
         )
-        if isinstance(result, (BaseModel, HttpxBinaryResponseContent)) and hasattr(result, "_hidden_params"):
+        if isinstance(priced_result, (BaseModel, HttpxBinaryResponseContent)) and hasattr(
+            priced_result, "_hidden_params"
+        ):
             hidden_params: Final = result_hidden_params
             if (
                 "response_cost" in hidden_params and hidden_params["response_cost"] is not None
@@ -1799,7 +1807,7 @@ class Logging(LiteLLMLoggingBaseClass):
 
         try:
             response_cost_calculator_kwargs: Final = {
-                "response_object": result,
+                "response_object": priced_result,
                 "model": litellm_model_name or self.model,
                 "cache_hit": cache_hit,
                 "custom_llm_provider": self.model_call_details.get("custom_llm_provider", None),

tests/integration/providers/test_fal_ai_video_wire.py

@@ -76,6 +76,9 @@ def test_fal_h3_video_create_uses_canonical_body_and_status_path(gateway: Gatewa
     request_id: Final = "fal-h3-req-" + uuid.uuid4().hex
 
     def respond(request: Request) -> Reply:
+        if request.target == f"/files/{request_id}.mp4":
+            assert request.method == "GET"
+            return Reply(body=_MP4, content_type="video/mp4")
         assert request.headers["authorization"] == "Key synthetic-fal-key"
         if request.method == "POST":
             assert request.target == f"/{_H3_MODEL}"

tests/integration/proxy_config.yaml

@@ -5,6 +5,7 @@ general_settings:
   store_model_in_db: true
   disable_spend_logs: false
   proxy_batch_write_at: 1
+  allow_unmanaged_response_ids: true
 litellm_settings:
   enable_redis_auth_cache: true
   cache: true

tests/test_litellm/litellm_core_utils/test_litellm_logging.py

@@ -7467,3 +7467,24 @@ def test_get_assembled_streaming_response_without_usage_cost_leaves_pricing_to_t
     assert "additional_headers" not in assembled._hidden_params
     price_map_cost = logging_obj._response_cost_calculator(result=assembled)
     assert price_map_cost is not None and 0 < price_map_cost != 0.0042
+
+
+def test_response_cost_calculator_prices_terminal_responses_event_from_its_response():
+    logging_obj: Final = _responses_stream_logging_obj()
+    inner_response: Final = ResponsesAPIResponse(
+        id="resp-priced",
+        created_at=1,
+        object="response",
+        status="completed",
+        model="gpt-4o-mini",
+        output=[],
+        usage=ResponseAPIUsage(input_tokens=1840, output_tokens=412, total_tokens=2252),
+    )
+    event: Final = ResponseCompletedEvent(type="response.completed", response=inner_response)
+
+    event_cost: Final = logging_obj._response_cost_calculator(result=event)
+    inner_cost: Final = logging_obj._response_cost_calculator(result=inner_response)
+
+    assert event_cost is not None and event_cost > 0
+    assert event_cost == inner_cost
+    assert logging_obj.cost_breakdown["input_cost"] is not None and logging_obj.cost_breakdown["input_cost"] > 0