fix(lowest_cost): resolve provider-prefixed model pricing via get_model_info

litellm/router_strategy/lowest_cost.py

@@ -246,6 +246,13 @@ class LowestCostLoggingHandler(CustomLogger):
             )
             item_litellm_model_name = _deployment.get("litellm_params", {}).get("model")
             item_litellm_model_cost_map = litellm.model_cost.get(item_litellm_model_name, {})
+            if not item_litellm_model_cost_map and item_litellm_model_name:
+                try:
+                    item_litellm_model_cost_map = dict(
+                        litellm.get_model_info(model=item_litellm_model_name)
+                    )
+                except Exception:
+                    item_litellm_model_cost_map = {}
 
             # check if user provided input_cost_per_token and output_cost_per_token in litellm_params
             item_input_cost = None

tests/local_testing/test_lowest_cost_routing.py

@@ -49,6 +49,36 @@ async def test_get_available_deployments():
     assert selected_model["model_info"]["id"] == "groq-llama"
 
 
+@pytest.mark.asyncio
+async def test_get_available_deployments_provider_prefixed_model_pricing():
+    test_cache = DualCache()
+    model_list = [
+        {
+            "model_name": "gpt-3.5-turbo",
+            "litellm_params": {"model": "openai/gpt-4o-mini"},
+            "model_info": {"id": "prefixed-openai"},
+        },
+        {
+            "model_name": "gpt-3.5-turbo",
+            "litellm_params": {
+                "model": "azure/some-custom-deployment",
+                "input_cost_per_token": 0.5,
+                "output_cost_per_token": 0.5,
+            },
+            "model_info": {"id": "custom-priced"},
+        },
+    ]
+    lowest_cost_logger = LowestCostLoggingHandler(
+        router_cache=test_cache,
+    )
+
+    selected_model = await lowest_cost_logger.async_get_available_deployments(
+        model_group="gpt-3.5-turbo", healthy_deployments=model_list
+    )
+
+    assert selected_model["model_info"]["id"] == "prefixed-openai"
+
+
 @pytest.mark.asyncio
 async def test_get_available_deployments_custom_price():
     from litellm._logging import verbose_router_logger