From 1a163895849951056c3a145900400ca1b2e22cd4 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E6=9E=97SO?=
 <142557582+Linxiushen@users.noreply.github.com>
Date: Wed, 5 Aug 2026 06:00:19 +0800
Subject: [PATCH] fix(router): validate static cost metadata

---
 litellm/router_strategy/lowest_cost.py        | 24 +++++++++++----
 .../router_strategy/test_lowest_cost.py       | 29 +++++++++++++++++++
 2 files changed, 47 insertions(+), 6 deletions(-)

diff --git a/litellm/router_strategy/lowest_cost.py b/litellm/router_strategy/lowest_cost.py
index 25069c5d803..e72685ddc89 100644
--- a/litellm/router_strategy/lowest_cost.py
+++ b/litellm/router_strategy/lowest_cost.py
@@ -1,8 +1,9 @@
 #### What this does ####
 #   picks based on response time (for streaming, this is time to first token)
 from datetime import datetime, timedelta
-from typing import Final, cast
+from typing import Final
 
+from pydantic import TypeAdapter, ValidationError
 from typing_extensions import TypedDict
 
 import litellm
@@ -18,14 +19,25 @@ class _ModelCostInfo(TypedDict, total=False):
     litellm_provider: str | None
 
 
+_MODEL_COST_INFO_ADAPTER: Final = TypeAdapter(_ModelCostInfo)
+
+
+def _get_validated_model_cost_info(model_name: str) -> _ModelCostInfo | None:
+    raw_model_cost_info: Final[object] = litellm.model_cost.get(model_name)
+    if raw_model_cost_info is None:
+        return None
+
+    try:
+        return _MODEL_COST_INFO_ADAPTER.validate_python(raw_model_cost_info)
+    except ValidationError:
+        return None
+
+
 def _get_model_cost_info(model_name: str | None) -> _ModelCostInfo:
     if model_name is None:
         return {}  # mutable-ok: each unresolved model gets an independent cost map
 
-    model_cost = cast(  # cast-ok: model_cost values come from the typed model cost JSON
-        dict[str, _ModelCostInfo], litellm.model_cost
-    )
-    exact_model_cost = model_cost.get(model_name)
+    exact_model_cost: Final = _get_validated_model_cost_info(model_name)
     if exact_model_cost is not None:
         return exact_model_cost
 
@@ -33,7 +45,7 @@ def _get_model_cost_info(model_name: str | None) -> _ModelCostInfo:
     if separator == "":
         return {}  # mutable-ok: each unresolved model gets an independent cost map
 
-    unprefixed_model_cost = model_cost.get(unprefixed_model_name)
+    unprefixed_model_cost: Final = _get_validated_model_cost_info(unprefixed_model_name)
     if unprefixed_model_cost is None or unprefixed_model_cost.get("litellm_provider") != provider_name:
         return {}
     return unprefixed_model_cost
diff --git a/tests/test_litellm/router_strategy/test_lowest_cost.py b/tests/test_litellm/router_strategy/test_lowest_cost.py
index 78bbda388ca..b426e5bbde1 100644
--- a/tests/test_litellm/router_strategy/test_lowest_cost.py
+++ b/tests/test_litellm/router_strategy/test_lowest_cost.py
@@ -57,3 +57,32 @@ async def test_unknown_provider_model_does_not_query_dynamic_metadata() -> None:
     assert selected is not None
     assert selected["model_info"]["id"] == "luna"
     get_model_info.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_invalid_static_cost_entry_is_ignored() -> None:
+    deployments = [
+        {
+            "model_name": "test-group",
+            "litellm_params": {"model": "test-provider/corrupt-cost"},
+            "model_info": {"id": "invalid"},
+        },
+        {
+            "model_name": "test-group",
+            "litellm_params": {"model": "openai/gpt-5.6-luna"},
+            "model_info": {"id": "luna"},
+        },
+    ]
+    handler = LowestCostLoggingHandler(router_cache=DualCache())
+
+    with patch.dict(
+        litellm.model_cost,
+        {"test-provider/corrupt-cost": {"input_cost_per_token": "invalid"}},
+    ):
+        selected = await handler.async_get_available_deployments(
+            model_group="test-group",
+            healthy_deployments=deployments,
+        )
+
+    assert selected is not None
+    assert selected["model_info"]["id"] == "luna"