fix(litellm): stamp cache-hit marker on HiddenParams carriers and restore probe side effects

gauntlet-mini on ace57d25f verified two live gaps. TextCompletionResponse and
ResponsesAPIResponse carry _hidden_params as a HiddenParams object, so the
dict-gated marker write skipped them and the served hit still stamped the full
price. The marker write is now container-agnostic, which it already was for the
partial-embedding stamp, and a regression test exercises the carrier end to end.
_uncached_response_cost also restores billed_token_rates, zero_cost_warned, and
the response_cost_failure_debug_information key alongside cost_breakdown, so the
probe leaves no pricing state on the call it reads.

litellm/caching/caching_handler.py

@@ -963,11 +963,7 @@ class LLMCachingHandler:
                 )
 
                 response_obj: Final = ResponsesAPIResponse(**cached_result)
-                if (
-                    hasattr(response_obj, "_hidden_params")
-                    and response_obj._hidden_params is not None
-                    and isinstance(response_obj._hidden_params, dict)
-                ):
+                if hasattr(response_obj, "_hidden_params") and response_obj._hidden_params is not None:
                     response_obj._hidden_params["cache_hit"] = True
 
                 if _stream_replay_requested(kwargs):
@@ -980,11 +976,7 @@ class LLMCachingHandler:
                 else:
                     cached_result = response_obj
 
-        if (
-            hasattr(cached_result, "_hidden_params")
-            and cached_result._hidden_params is not None
-            and isinstance(cached_result._hidden_params, dict)
-        ):
+        if hasattr(cached_result, "_hidden_params") and cached_result._hidden_params is not None:
             cached_result._hidden_params["cache_hit"] = True
 
         #########################################################

litellm/litellm_core_utils/litellm_logging.py

@@ -1976,12 +1976,23 @@ class Logging(LiteLLMLoggingBaseClass):
             list,
         ],
     ) -> float | None:
-        """Price a response as if it had missed the cache, without overwriting this call's cost breakdown"""
+        """Price a response as if it had missed the cache, without leaving the probe's side effects on this call"""
         previous_cost_breakdown: Final = self.cost_breakdown
+        previous_billed_token_rates: Final = self.billed_token_rates
+        previous_zero_cost_warned: Final = self.zero_cost_warned
+        debug_key: Final = "response_cost_failure_debug_information"
+        debug_missing: Final = object()
+        debug_before: Final = self.model_call_details.get(debug_key, debug_missing)
         try:
             return self._response_cost_calculator(result=result, cache_hit=False, ignore_stamped_cost=True)
         finally:
             self.cost_breakdown = previous_cost_breakdown
+            self.billed_token_rates = previous_billed_token_rates
+            self.zero_cost_warned = previous_zero_cost_warned
+            if debug_before is debug_missing:
+                self.model_call_details.pop(debug_key, None)
+            else:
+                self.model_call_details[debug_key] = debug_before
 
     def _record_zero_cost_diagnostic(
         self,

tests/unit/litellm_core_utils/llm_response_utils/test_response_metadata.py

@@ -26,7 +26,7 @@ from litellm.litellm_core_utils.llm_response_utils.response_metadata import (
 )
 from litellm.proxy._types import UserAPIKeyAuth
 from litellm.proxy.common_request_processing import ProxyBaseLLMRequestProcessing
-from litellm.types.utils import ModelResponse, Usage
+from litellm.types.utils import ModelResponse, TextCompletionResponse, Usage
 
 
 class TestCallbackDurationMs:
@@ -703,3 +703,60 @@ def test_update_response_metadata_stamps_zero_response_cost_for_cache_hit_result
 
     assert result._hidden_params["response_cost"] == 0.0
     assert logging_obj.cost_breakdown is None
+
+
+def test_update_response_metadata_stamps_zero_response_cost_for_cache_hit_hidden_params_carrier(monkeypatch):
+    """A cache hit whose _hidden_params is a HiddenParams object (text_completion's carrier) stamps 0 cost, not the full price."""
+    monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True")
+    monkeypatch.setattr(litellm, "model_cost", litellm.get_model_cost_map(url=""))
+
+    deployment_id: Final = "cache-hit-deployment-hiddenparams"
+    litellm.register_model(
+        model_cost={
+            deployment_id: {
+                "input_cost_per_token": 1e-06,
+                "output_cost_per_token": 2e-06,
+                "litellm_provider": "openai",
+                "mode": "completion",
+            }
+        }
+    )
+    start_time: Final = datetime.datetime(2026, 9, 28, 12, 0, 0)
+    logging_obj: Final = Logging(
+        model="gpt-3.5-turbo-instruct",
+        messages="Hi",
+        stream=False,
+        call_type="text_completion",
+        start_time=start_time,
+        litellm_call_id="cache-hit-hiddenparams",
+        function_id="f",
+    )
+    logging_obj.update_environment_variables(
+        model="gpt-3.5-turbo-instruct",
+        litellm_params={
+            "metadata": {"model_info": {"id": deployment_id}},
+        },
+        optional_params={},
+        custom_llm_provider="openai",
+    )
+    result: Final = TextCompletionResponse(
+        id="cmpl-x",
+        object="text_completion",
+        created=1,
+        model="gpt-3.5-turbo-instruct",
+        choices=[{"text": "ok", "index": 0, "finish_reason": "stop"}],
+        usage={"prompt_tokens": 11, "completion_tokens": 7, "total_tokens": 18},
+    )
+    result._hidden_params["cache_hit"] = True
+
+    update_response_metadata(
+        result=result,
+        logging_obj=logging_obj,
+        model="gpt-3.5-turbo-instruct",
+        kwargs={"model_info": {"id": deployment_id}},
+        start_time=start_time,
+        end_time=start_time + datetime.timedelta(seconds=2),
+    )
+
+    assert result._hidden_params.get("response_cost") == 0.0
+    assert logging_obj.cost_breakdown is None