From 99cd54f1544b5cd7e964e37bd431cfbfdb9ca2c5 Mon Sep 17 00:00:00 2001
From: Surender Gulia <shriomgulia5@gmail.com>
Date: Thu, 5 Nov 2026 05:01:20 +0000
Subject: [PATCH] fix(proxy/vertex_ai): update token handling logic

---
 0                                           | 0
 tests/unit/proxy/test_budget_reservation.py | 9 +++++++++
 2 files changed, 9 insertions(+)
 create mode 100644 0

diff --git a/0 b/0
new file mode 100644
index 00000000000..e69de29bb2d
diff --git a/tests/unit/proxy/test_budget_reservation.py b/tests/unit/proxy/test_budget_reservation.py
index 6ce8d400daf..81df546e94a 100644
--- a/tests/unit/proxy/test_budget_reservation.py
+++ b/tests/unit/proxy/test_budget_reservation.py
@@ -3721,3 +3721,12 @@ async def test_unreserved_model_access_group_is_charged_alongside_a_reserved_one
     assert counter_cache.in_memory_cache.get_cache(
         key=model_access_group_spend_counter_key("starter")
     ) == pytest.approx(4.2)
+
+def test_requested_output_tokens_gemini_generation_config() -> None:
+    """Verify generationConfig.maxOutputTokens is properly parsed for Gemini models."""
+    from litellm.proxy.spend_tracking.budget_reservation import _requested_output_tokens
+
+    body = {
+        "generationConfig": {"maxOutputTokens": 4096}
+    }
+    assert _requested_output_tokens(body) == 4096
\ No newline at end of file