fix(parallel_ai): stop a caller from pricing its own search request

`_parallel_ai_usage` carries the provider's reported usage into cost
calculation. It was only written when the response contained a usage block, so
a caller could pass `_parallel_ai_usage=[{"name": "sku_search", "count": 0}]`
and, whenever the provider omitted usage, bill $0.00 instead of $0.005 — the
value also reached the upstream request body as an unknown field.

The key is now stripped from inbound params and written unconditionally from
the parsed response, so only the provider can populate it.

litellm/llms/parallel_ai/search/transformation.py

@@ -214,6 +214,10 @@ class ParallelAISearchConfig(BaseSearchConfig):
         # unified-spec param with no v1 equivalent
         params.pop("max_tokens_per_page", None)
 
+        # reserved for the provider's own reported usage, which prices the request;
+        # a caller-supplied value would otherwise set its own cost
+        params.pop(PARALLEL_AI_USAGE_PARAM, None)
+
         return {**request_data, **params}
 
     def transform_search_response(
@@ -237,11 +241,12 @@ class ParallelAISearchConfig(BaseSearchConfig):
         """
         parsed: Final = _ParallelAIV1SearchResponse.model_validate(raw_response.json())
 
-        if parsed.usage is not None:
-            logging_obj.optional_params = {
-                **logging_obj.optional_params,
-                PARALLEL_AI_USAGE_PARAM: parsed.usage,
-            }
+        # written unconditionally: leaving a caller-supplied value in place when the
+        # provider reports no usage would let the caller price its own request
+        logging_obj.optional_params = {
+            **logging_obj.optional_params,
+            PARALLEL_AI_USAGE_PARAM: parsed.usage,
+        }
 
         results: Final = tuple(
             SearchResult.model_validate(

tests/test_litellm/llms/parallel_ai/test_parallel_ai_search.py

@@ -472,3 +472,27 @@ class TestParallelAISearch:
             )
 
             assert response._hidden_params["response_cost"] == pytest.approx(0.005)
+
+    @pytest.mark.asyncio
+    async def test_caller_cannot_supply_provider_usage(self, bundled_cost_map):
+        """`_parallel_ai_usage` prices the request, so a caller must not be able to set it.
+
+        The provider reports no usage here, which is the case where a caller-supplied
+        value would otherwise survive into the cost calculation.
+        """
+        response_payload = {k: v for k, v in MOCK_V1_RESPONSE.items() if k != "usage"}
+        with patch(
+            "litellm.llms.custom_httpx.http_handler.AsyncHTTPHandler.post",
+            new_callable=AsyncMock,
+        ) as mock_post:
+            mock_post.return_value = _mock_response(response_payload)
+
+            response = await litellm.asearch(
+                query="AI developments",
+                search_provider="parallel_ai",
+                mode="basic",
+                _parallel_ai_usage=[{"name": "sku_search", "count": 0}],
+            )
+
+            assert response._hidden_params["response_cost"] == pytest.approx(0.005)
+            assert "_parallel_ai_usage" not in mock_post.call_args.kwargs["json"]