From 68d19ef63a0794bc44ed03e95cd9b6998915de01 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=86=AF=E5=9F=BA=E9=AD=81?= <1412414664@qq.com>
Date: Tue, 23 Jun 2026 02:21:07 +0800
Subject: [PATCH] fix(proxy): skip personal budget for zero-cost models

---
 litellm/proxy/hooks/max_budget_limiter.py     | 12 ++-
 .../proxy/hooks/test_max_budget_limiter.py    | 80 +++++++++++++++++++
 2 files changed, 91 insertions(+), 1 deletion(-)

diff --git a/litellm/proxy/hooks/max_budget_limiter.py b/litellm/proxy/hooks/max_budget_limiter.py
index 9a7e5117945..7bd8a226315 100644
--- a/litellm/proxy/hooks/max_budget_limiter.py
+++ b/litellm/proxy/hooks/max_budget_limiter.py
@@ -47,7 +47,17 @@ class _PROXY_MaxBudgetLimiter(CustomLogger):
             ):
                 return
 
-            from litellm.proxy.proxy_server import get_current_spend
+            from litellm.proxy.auth.auth_checks import _is_model_cost_zero
+            from litellm.proxy.proxy_server import get_current_spend, llm_router
+
+            if data and _is_model_cost_zero(
+                model=data.get("model"), llm_router=llm_router
+            ):
+                verbose_proxy_logger.info(
+                    "MaxBudgetLimiter: skipping personal budget for zero-cost model=%s",
+                    data.get("model"),
+                )
+                return
 
             curr_spend = await get_current_spend(
                 counter_key=user_counter_key,
diff --git a/tests/test_litellm/proxy/hooks/test_max_budget_limiter.py b/tests/test_litellm/proxy/hooks/test_max_budget_limiter.py
index 0074d7062b8..dd62f41a19a 100644
--- a/tests/test_litellm/proxy/hooks/test_max_budget_limiter.py
+++ b/tests/test_litellm/proxy/hooks/test_max_budget_limiter.py
@@ -19,6 +19,7 @@ from fastapi import HTTPException
 from litellm.caching.caching import DualCache
 from litellm.proxy._types import UserAPIKeyAuth
 from litellm.proxy.hooks.max_budget_limiter import _PROXY_MaxBudgetLimiter
+from litellm.router import Router
 
 
 def _make_user_api_key_auth(
@@ -185,6 +186,85 @@ async def test_team_keys_skip_personal_budget():
     mock_get_spend.assert_not_awaited()
 
 
+@pytest.mark.asyncio
+async def test_zero_cost_model_skips_personal_budget():
+    handler = _PROXY_MaxBudgetLimiter()
+    user_api_key_dict = _make_user_api_key_auth(
+        user_id="user-1",
+        user_max_budget=10.0,
+        user_spend=99.0,
+    )
+    router = Router(
+        model_list=[
+            {
+                "model_name": "free-model",
+                "litellm_params": {
+                    "model": "openai/free-model",
+                    "api_key": "sk-test",
+                    "input_cost_per_token": 0.0,
+                    "output_cost_per_token": 0.0,
+                },
+            }
+        ]
+    )
+
+    with (
+        patch("litellm.proxy.proxy_server.llm_router", router),
+        patch(
+            "litellm.proxy.proxy_server.get_current_spend",
+            new=AsyncMock(return_value=99.0),
+        ) as mock_get_spend,
+    ):
+        result = await handler.async_pre_call_hook(
+            user_api_key_dict=user_api_key_dict,
+            cache=DualCache(),
+            data={"model": "free-model"},
+            call_type="completion",
+        )
+
+    assert result is None
+    mock_get_spend.assert_not_awaited()
+
+
+@pytest.mark.asyncio
+async def test_paid_model_still_enforces_personal_budget():
+    handler = _PROXY_MaxBudgetLimiter()
+    user_api_key_dict = _make_user_api_key_auth(
+        user_id="user-1",
+        user_max_budget=10.0,
+        user_spend=99.0,
+    )
+    router = Router(
+        model_list=[
+            {
+                "model_name": "paid-model",
+                "litellm_params": {
+                    "model": "openai/gpt-4o-mini",
+                    "api_key": "sk-test",
+                },
+            }
+        ]
+    )
+
+    with (
+        patch("litellm.proxy.proxy_server.llm_router", router),
+        patch(
+            "litellm.proxy.proxy_server.get_current_spend",
+            new=AsyncMock(return_value=99.0),
+        ),
+    ):
+        with pytest.raises(HTTPException) as exc_info:
+            await handler.async_pre_call_hook(
+                user_api_key_dict=user_api_key_dict,
+                cache=DualCache(),
+                data={"model": "paid-model"},
+                call_type="completion",
+            )
+
+    assert exc_info.value.status_code == 429
+    assert "Max budget limit reached." in exc_info.value.detail
+
+
 @pytest.mark.asyncio
 async def test_no_max_budget_passes():
     handler = _PROXY_MaxBudgetLimiter()