Merge 0e158d1ba5 into e768ad55ce

litellm/litellm_core_utils/litellm_logging.py

@@ -2404,7 +2404,7 @@ class Logging(LiteLLMLoggingBaseClass):
 
         if self.model_call_details.get("cache_hit") is True:
             self.model_call_details["response_cost"] = 0.0
-        elif "response_cost" in hidden_params:
+        elif hidden_params.get("response_cost") is not None:
             self.model_call_details["response_cost"] = hidden_params["response_cost"]
             self._record_zero_cost_diagnostic(logging_result, hidden_params["response_cost"])
         elif (existing_cost := self.model_call_details.get("response_cost")) is not None and existing_cost != 0:
@@ -2413,7 +2413,16 @@ class Logging(LiteLLMLoggingBaseClass):
             # Do not preserve 0 from failure_handler on intermediate router retries.
             pass
         else:
-            self.model_call_details["response_cost"] = self._response_cost_calculator(result=logging_result)
+            if "response_cost" in hidden_params:
+                warned = self.zero_cost_warned
+                self.zero_cost_warned = True
+                try:
+                    self.model_call_details["response_cost"] = self._response_cost_calculator(result=logging_result)
+                finally:
+                    self.model_call_details["zero_cost_diagnostic"] = None
+                    self.zero_cost_warned = warned
+            else:
+                self.model_call_details["response_cost"] = self._response_cost_calculator(result=logging_result)
 
         if not build_logging_payload:
             return

tests/unit/litellm_core_utils/test_litellm_logging.py

@@ -4296,6 +4296,39 @@ def test_process_hidden_params_uses_hidden_params_cost_after_failure_handler_zer
     assert slo.get("response_cost") == passthrough_cost
 
 
+def test_process_hidden_params_falls_through_on_none_response_cost():
+    from datetime import datetime
+
+    from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
+    from litellm.types.utils import ModelResponse, Usage
+
+    logging_obj = LiteLLMLoggingObj(
+        model="openai/gpt-4o-mini",
+        messages=[{"role": "user", "content": "hi"}],
+        stream=False,
+        call_type="responses",
+        start_time=datetime.now(),
+        litellm_call_id="test-none-hidden-cost",
+        function_id="test-none-hidden-cost",
+    )
+    logging_obj.model_call_details["litellm_params"] = {"model": "openai/gpt-4o-mini"}
+    logging_obj.optional_params = {}
+
+    result = ModelResponse(
+        id="success",
+        choices=[{"message": {"role": "assistant", "content": "ok"}}],
+        usage=Usage(prompt_tokens=100, completion_tokens=10, total_tokens=110),
+    )
+    result._hidden_params = {"response_cost": None}
+
+    logging_obj._process_hidden_params_and_response_cost(result, datetime.now(), datetime.now())
+
+    cost = logging_obj.model_call_details.get("response_cost")
+    assert cost is not None and cost > 0
+    slo = logging_obj.model_call_details.get("standard_logging_object") or {}
+    assert slo.get("response_cost", 0) > 0
+
+
 def test_function_setup_litellm_metadata_populates_metadata():
     """
     Test that function_setup() properly handles litellm_metadata (used by /v1/messages,