Merge pull request #36397 from ousamabenyounes/litellm_fix_gemini_web_search_unique_queries_36377

fix(vertex_ai): bill Gemini grounding per unique web search query

litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py

@@ -1979,16 +1979,15 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig):
 
     @staticmethod
     def _calculate_web_search_requests(grounding_metadata: list[dict]) -> int | None:
-        web_search_requests: int | None = None
-
-        if grounding_metadata and isinstance(grounding_metadata, list) and len(grounding_metadata) > 0:
-            for grounding_metadata_item in grounding_metadata:
-                web_search_queries = grounding_metadata_item.get("webSearchQueries")
-                if web_search_queries and web_search_requests:
-                    web_search_requests += len([q for q in web_search_queries if q])
-                elif web_search_queries:
-                    web_search_requests = len([q for q in web_search_queries if q])
-        return web_search_requests
+        if not (grounding_metadata and isinstance(grounding_metadata, list)):
+            return None
+        unique_queries: Final = {
+            query
+            for grounding_metadata_item in grounding_metadata
+            for query in (grounding_metadata_item.get("webSearchQueries") or [])
+            if query
+        }
+        return len(unique_queries) or None
 
     @staticmethod
     def _create_streaming_choice(

tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py

@@ -2,7 +2,7 @@ import asyncio
 import json
 import re
 from copy import deepcopy
-from typing import List, cast
+from typing import Final, List, cast
 from unittest.mock import MagicMock, patch
 
 import pytest
@@ -5579,3 +5579,25 @@ def test_accumulated_json_async_end_of_stream_drains_buffered_value():
     result = asyncio.run(iterator.__anext__())
     assert result is not None
     assert result.choices[0].delta.content == "a"
+
+
+def test_calculate_web_search_requests_counts_unique_queries():
+    """Gemini 3 per_query billing charges per unique query executed, not per emitted string.
+
+    Regression for #36377: duplicate webSearchQueries within and across grounding
+    metadata items must collapse to the distinct-query count, and empty strings must
+    be ignored, matching Google's documented Grounding-with-Search billing rule.
+    """
+    duplicates_in_one_item: Final = [
+        {"webSearchQueries": ["euro 2024 winner", "euro 2024 winner", "spain england final", ""]}
+    ]
+    assert VertexGeminiConfig._calculate_web_search_requests(duplicates_in_one_item) == 2
+
+    duplicates_across_items: Final = [
+        {"webSearchQueries": ["euro 2024 winner"]},
+        {"webSearchQueries": ["euro 2024 winner", "spain england final"]},
+    ]
+    assert VertexGeminiConfig._calculate_web_search_requests(duplicates_across_items) == 2
+
+    assert VertexGeminiConfig._calculate_web_search_requests([]) is None
+    assert VertexGeminiConfig._calculate_web_search_requests([{"webSearchQueries": ["", ""]}]) is None