From e06ca6e1eec00e2a471a94dd1f918d9fd1135a46 Mon Sep 17 00:00:00 2001
From: Anuj7411 <ojhaanuj224@gmail.com>
Date: Wed, 26 Aug 2026 19:28:49 +0530
Subject: [PATCH] fix(vertex): bill above-128k character output by characters,
 not tokens

---
 litellm/llms/vertex_ai/cost_calculator.py  |  3 +-
 tests/test_litellm/test_cost_calculator.py | 44 ++++++++++++++++++++++
 2 files changed, 45 insertions(+), 2 deletions(-)

diff --git a/litellm/llms/vertex_ai/cost_calculator.py b/litellm/llms/vertex_ai/cost_calculator.py
index 23cb1e5b580..5362b8597df 100644
--- a/litellm/llms/vertex_ai/cost_calculator.py
+++ b/litellm/llms/vertex_ai/cost_calculator.py
@@ -133,7 +133,6 @@ def cost_per_character(
             usage=usage,
         )
     else:
-        completion_tokens: Final = usage.completion_tokens
         try:
             if (
                 _is_above_128k(tokens=completion_characters * 4)  # 1 token = 4 char
@@ -145,7 +144,7 @@ def cost_per_character(
                 ), (
                     f"model info for model={model} does not have 'output_cost_per_character_above_128k_tokens' pricing\nmodel_info={model_info}"
                 )
-                completion_cost = completion_tokens * model_info["output_cost_per_character_above_128k_tokens"]
+                completion_cost = completion_characters * model_info["output_cost_per_character_above_128k_tokens"]
             else:
                 assert (
                     "output_cost_per_character" in model_info and model_info["output_cost_per_character"] is not None
diff --git a/tests/test_litellm/test_cost_calculator.py b/tests/test_litellm/test_cost_calculator.py
index 98938dee62e..5ddc64ca756 100644
--- a/tests/test_litellm/test_cost_calculator.py
+++ b/tests/test_litellm/test_cost_calculator.py
@@ -1817,6 +1817,50 @@ def test_vertex_uplift_composes_with_above_128k_pricing(monkeypatch):
     assert regional_completion == pytest.approx(global_completion * 1.10, rel=1e-9)
 
 
+def test_vertex_above_128k_output_bills_per_character_not_per_token(monkeypatch):
+    """Regression: the above-128k output branch of the character-priced Vertex path
+    must multiply the per-character rate by the completion character count, not the
+    token count. Token counts run ~4x smaller than character counts, so the bug
+    undercharged large completions by roughly that factor."""
+    from litellm.llms.vertex_ai.cost_calculator import cost_per_character
+
+    model = "fake-char-128k-model"
+    monkeypatch.setattr(
+        litellm,
+        "model_cost",
+        {
+            **litellm.get_model_cost_map(url=""),
+            f"vertex_ai/{model}": {
+                "litellm_provider": "vertex_ai",
+                "mode": "chat",
+                "input_cost_per_character": 1e-07,
+                "output_cost_per_character": 2e-07,
+                "input_cost_per_character_above_128k_tokens": 2e-07,
+                "output_cost_per_character_above_128k_tokens": 4e-07,
+            },
+        },
+    )
+
+    completion_characters = 600_000  # > 512k -> above the 128k-token equivalent
+    completion_tokens = 150_000
+    usage = Usage(
+        prompt_tokens=10,
+        completion_tokens=completion_tokens,
+        total_tokens=10 + completion_tokens,
+    )
+
+    _, completion_cost_usd = cost_per_character(
+        model=model,
+        custom_llm_provider="vertex_ai",
+        usage=usage,
+        prompt_characters=100,
+        completion_characters=completion_characters,
+    )
+
+    assert completion_cost_usd == pytest.approx(completion_characters * 4e-07, rel=1e-9)
+    assert completion_cost_usd != pytest.approx(completion_tokens * 4e-07, rel=1e-9)
+
+
 def test_cost_discount_vertex_ai():
     """
     Test that cost discount is applied correctly for Vertex AI provider