fix(router): avoid dynamic cost metadata lookup

litellm/router_strategy/lowest_cost.py

@@ -1,7 +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, TypedDict, cast
+from typing import Final, cast
+
+from typing_extensions import TypedDict
 
 import litellm
 from litellm import ModelResponse, token_counter, verbose_logger
@@ -13,6 +15,7 @@ from litellm.integrations.custom_logger import CustomLogger
 class _ModelCostInfo(TypedDict, total=False):
     input_cost_per_token: float | None
     output_cost_per_token: float | None
+    litellm_provider: str | None
 
 
 def _get_model_cost_info(model_name: str | None) -> _ModelCostInfo:
@@ -26,13 +29,15 @@ def _get_model_cost_info(model_name: str | None) -> _ModelCostInfo:
     if exact_model_cost is not None:
         return exact_model_cost
 
-    try:
-        return cast(  # cast-ok: only the two declared numeric fields are read below
-            _ModelCostInfo, litellm.get_model_info(model_name)
-        )
-    except Exception:
+    provider_name, separator, unprefixed_model_name = model_name.partition("/")
+    if separator == "":
         return {}  # mutable-ok: each unresolved model gets an independent cost map
 
+    unprefixed_model_cost = model_cost.get(unprefixed_model_name)
+    if unprefixed_model_cost is None or unprefixed_model_cost.get("litellm_provider") != provider_name:
+        return {}
+    return unprefixed_model_cost
+
 
 class LowestCostLoggingHandler(CustomLogger):
     test_flag: bool = False

tests/test_litellm/router_strategy/test_lowest_cost.py

@@ -1,3 +1,6 @@
+from unittest.mock import patch
+
+import litellm
 import pytest
 
 from litellm.caching.caching import DualCache
@@ -27,3 +30,30 @@ async def test_provider_prefixed_models_use_resolved_costs() -> None:
 
     assert selected is not None
     assert selected["model_info"]["id"] == "luna"
+
+
+@pytest.mark.asyncio
+async def test_unknown_provider_model_does_not_query_dynamic_metadata() -> None:
+    deployments = [
+        {
+            "model_name": "test-group",
+            "litellm_params": {"model": "ollama/caller-controlled-model"},
+            "model_info": {"id": "unknown"},
+        },
+        {
+            "model_name": "test-group",
+            "litellm_params": {"model": "openai/gpt-5.6-luna"},
+            "model_info": {"id": "luna"},
+        },
+    ]
+    handler = LowestCostLoggingHandler(router_cache=DualCache())
+
+    with patch.object(litellm, "get_model_info") as get_model_info:
+        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"
+    get_model_info.assert_not_called()