[Fix] Unify cost calc in success_handler dict and typed branches (#26629)

* Unify cost calc in success_handler dict and typed branches

* Trim verbose comments and docstrings

---------

Co-authored-by: Michael Riad Zaky <michaelr@Mac.localdomain>
Co-authored-by: Michael Riad Zaky <michaelr@Michaels-MacBook-Air.local>

Author: 3 people

litellm/litellm_core_utils/litellm_logging.py

@@ -1467,6 +1467,8 @@ def _response_cost_calculator(
             LiteLLMRealtimeStreamLoggingObject,
             OpenAIModerationResponse,
             "SearchResponse",
+            dict,
+            list,
         ],
         cache_hit: Optional[bool] = None,
         litellm_model_name: Optional[str] = None,
@@ -1744,6 +1746,7 @@ def _process_hidden_params_and_response_cost(
         start_time,
         end_time,
     ):
+        """Resolve hidden params, compute response cost, and emit the standard logging payload."""
         hidden_params = getattr(logging_result, "_hidden_params", {})
         if hidden_params:
             if self.model_call_details.get("litellm_params") is not None:
@@ -1877,24 +1880,12 @@ def _success_handler_helper_fn(
             ):
                 if self._is_recognized_call_type_for_logging(
                     logging_result=logging_result
-                ):
+                ) or isinstance(logging_result, (dict, list)):
                     self._process_hidden_params_and_response_cost(
                         logging_result=logging_result,
                         start_time=start_time,
                         end_time=end_time,
                     )
-                elif isinstance(result, dict) or isinstance(result, list):
-                    self.model_call_details["standard_logging_object"] = (
-                        self._build_standard_logging_payload(
-                            result, start_time, end_time
-                        )
-                    )
-                    if (
-                        standard_logging_payload := self.model_call_details.get(
-                            "standard_logging_object"
-                        )
-                    ) is not None:
-                        emit_standard_logging_payload(standard_logging_payload)
             elif standard_logging_object is not None:
                 self.model_call_details["standard_logging_object"] = (
                     standard_logging_object

tests/test_litellm/litellm_core_utils/test_litellm_logging.py

@@ -2534,3 +2534,141 @@ def test_get_standard_logging_object_payload_includes_litellm_call_id(logging_ob
 
     assert payload is not None
     assert payload["litellm_call_id"] == call_id
+
+
+def _make_dict_logging_obj():
+    """Build a Logging instance configured for a non-streaming dict result."""
+    obj = LitellmLogging(
+        model="claude-haiku-4-5@20251001",
+        messages=[{"role": "user", "content": "hi"}],
+        stream=False,
+        call_type="acompletion",
+        litellm_call_id="test-call-id",
+        start_time=time.time(),
+        function_id="test-fn",
+    )
+    obj.model_call_details = {
+        "model": "claude-haiku-4-5@20251001",
+        "custom_llm_provider": "vertex_ai",
+        "litellm_params": {"metadata": {}},
+        "response_cost": None,
+    }
+    return obj
+
+
+def test_success_handler_computes_cost_for_dict_response():
+    """Non-streaming dict responses run through the cost calculator."""
+    logging_obj = _make_dict_logging_obj()
+    expected_cost = 0.42
+    with (
+        patch.object(
+            logging_obj,
+            "_response_cost_calculator",
+            return_value=expected_cost,
+        ) as mock_calc,
+        patch.object(
+            logging_obj,
+            "_build_standard_logging_payload",
+            return_value={"response_cost": expected_cost},
+        ),
+        patch(
+            "litellm.litellm_core_utils.litellm_logging.emit_standard_logging_payload"
+        ),
+        patch.object(
+            logging_obj,
+            "_is_recognized_call_type_for_logging",
+            return_value=False,
+        ),
+        patch.object(
+            logging_obj,
+            "_transform_usage_objects",
+            side_effect=lambda result: result,
+        ),
+    ):
+        logging_obj.success_handler(
+            result={"id": "msg_1"},
+            start_time=time.time(),
+            end_time=time.time(),
+        )
+        mock_calc.assert_called_once()
+        assert logging_obj.model_call_details["response_cost"] == expected_cost
+
+
+def test_success_handler_preserves_precomputed_cost_for_dict_response():
+    """Precomputed response_cost on model_call_details must not be overwritten."""
+    logging_obj = _make_dict_logging_obj()
+    precomputed_cost = 1.23
+    logging_obj.model_call_details["response_cost"] = precomputed_cost
+    with (
+        patch.object(
+            logging_obj,
+            "_response_cost_calculator",
+            return_value=9.99,
+        ) as mock_calc,
+        patch.object(
+            logging_obj,
+            "_build_standard_logging_payload",
+            return_value={"response_cost": precomputed_cost},
+        ),
+        patch(
+            "litellm.litellm_core_utils.litellm_logging.emit_standard_logging_payload"
+        ),
+        patch.object(
+            logging_obj,
+            "_is_recognized_call_type_for_logging",
+            return_value=False,
+        ),
+        patch.object(
+            logging_obj,
+            "_transform_usage_objects",
+            side_effect=lambda result: result,
+        ),
+    ):
+        logging_obj.success_handler(
+            result={"id": "msg_2"},
+            start_time=time.time(),
+            end_time=time.time(),
+        )
+        mock_calc.assert_not_called()
+        assert logging_obj.model_call_details["response_cost"] == precomputed_cost
+
+
+def test_success_handler_unified_helper_runs_for_typed_results():
+    """Recognized typed responses still flow through the unified helper."""
+    logging_obj = _make_dict_logging_obj()
+    expected_cost = 0.10
+    typed_result = MagicMock()
+    typed_result._hidden_params = {}
+
+    with (
+        patch.object(
+            logging_obj,
+            "_response_cost_calculator",
+            return_value=expected_cost,
+        ) as mock_calc,
+        patch.object(
+            logging_obj,
+            "_build_standard_logging_payload",
+            return_value={"response_cost": expected_cost},
+        ),
+        patch(
+            "litellm.litellm_core_utils.litellm_logging.emit_standard_logging_payload"
+        ),
+        patch.object(
+            logging_obj,
+            "_is_recognized_call_type_for_logging",
+            return_value=True,
+        ),
+        patch.object(
+            logging_obj,
+            "_transform_usage_objects",
+            side_effect=lambda result: result,
+        ),
+    ):
+        logging_obj.success_handler(
+            result=typed_result,
+            start_time=time.time(),
+            end_time=time.time(),
+        )
+        mock_calc.assert_called_once()
+        assert logging_obj.model_call_details["response_cost"] == expected_cost