fix(proxy): show no price on a wildcard row whose calls go to another deployment

litellm/proxy/auth/model_checks.py

@@ -565,28 +565,22 @@ def wildcard_filled_model(model_name_pattern: str, litellm_model: str, listed_na
     )
 
 
-def _deployment_routed_model(source: Mapping[str, object], listed_name: str) -> str | None:
-    source_pattern: Final = source.get("model_name")
-    source_model: Final = _litellm_model(source)
-    return (
-        wildcard_filled_model(source_pattern, source_model, listed_name)
-        if isinstance(source_pattern, str) and isinstance(source_model, str)
-        else None
-    )
-
-
-def wildcard_routed_model(row: Mapping[str, object], llm_router: Router | None) -> str | None:
+def wildcard_row_priced_model(row: Mapping[str, object], llm_router: Router | None) -> str | None:
     model_id: Final = _model_info_field(row, "id")
     listed_name: Final = row.get("model_name")
     if llm_router is None or not isinstance(model_id, str) or not isinstance(listed_name, str):
         return None
     registry: Final = llm_router.registry()
     source: Final = _router_deployment(registry, model_id)
-    if source is None or _litellm_model(row) == _litellm_model(source):
+    if source is None:
+        return None
+    source_pattern: Final = source.get("model_name")
+    source_model: Final = _litellm_model(source)
+    if not isinstance(source_pattern, str) or not isinstance(source_model, str) or _litellm_model(row) == source_model:
         return None
     if model_id not in registry.names(listed_name, kind="route_candidates", team_id=_deployment_team_id(row)):
-        return None
-    return _deployment_routed_model(source, listed_name)
+        return source_model
+    return wildcard_filled_model(source_pattern, source_model, listed_name) or source_model
 
 
 def _get_wildcard_models(

litellm/proxy/proxy_server.py

@@ -443,7 +443,7 @@ from litellm.proxy.auth.model_checks import (
     get_key_models,
     get_mcp_server_ids,
     get_team_models,
-    wildcard_routed_model,
+    wildcard_row_priced_model,
 )
 from litellm.proxy.auth.password_policy import validate_password_not_breached, validate_password_policy
 from litellm.proxy.auth.user_api_key_auth import (  # noqa: F401, RUF100  # legacy module exports
@@ -16145,11 +16145,9 @@ def _enrich_model_info_with_litellm_data(
 
     # read litellm model_prices_and_context_window.json to get the following:
     # input_cost_per_token, output_cost_per_token, max_tokens
-    routed_model: Final = wildcard_routed_model(model, llm_router)
+    priced_as: Final = wildcard_row_priced_model(model, llm_router)
     priced_model: Final = (
-        model
-        if routed_model is None
-        else {**model, "litellm_params": {**model["litellm_params"], "model": routed_model}}
+        model if priced_as is None else {**model, "litellm_params": {**model["litellm_params"], "model": priced_as}}
     )
     litellm_model_info = get_litellm_model_info(model=priced_model)
 

tests/unit/proxy/proxy_server/test_routes_model_info.py

@@ -678,8 +678,9 @@ def test_model_info_never_prices_a_wildcard_row_as_a_route_its_calls_skip(
 ) -> None:
     """``acme-lite`` has a deployment of its own, ``acme-pro`` is an alias of it, ``acme-max`` is a
     routing group over it, and ``my-openai/gpt-acme-*`` is a narrower wildcard, so calls to those names
-    never reach the catch-all or ``my-openai/*``, which still list them. Such a row may go unpriced, but
-    any price a row shows must be the one its calls are billed at."""
+    never reach the catch-all or ``my-openai/*``, which still list them. Such a row shows no price, even
+    where another provider's model goes by the same name, and every price shown is the one calls are
+    billed at."""
     from litellm.proxy.auth import model_checks
 
     catalog: Final = {
@@ -713,6 +714,18 @@ def test_model_info_never_prices_a_wildcard_row_as_a_route_its_calls_skip(
             "litellm_provider": "openai",
             "mode": "chat",
         },
+        "acme-pro": {
+            "input_cost_per_token": 9e-07,
+            "output_cost_per_token": 1e-06,
+            "litellm_provider": "deepseek",
+            "mode": "chat",
+        },
+        "acme-max": {
+            "input_cost_per_token": 1.1e-06,
+            "output_cost_per_token": 1.2e-06,
+            "litellm_provider": "deepseek",
+            "mode": "chat",
+        },
     }
     monkeypatch.setattr(litellm, "model_cost", {**litellm.model_cost, **catalog})
     _invalidate_model_cost_lowercase_map()
@@ -779,6 +792,12 @@ def test_model_info_never_prices_a_wildcard_row_as_a_route_its_calls_skip(
         ("openai-wildcard", "my-openai/gpt-acme-lite"),
     ]
     assert listed[("deepseek-acme-lite", "acme-lite")] == (7e-07, 8e-07)
+    assert {key: prices for key, prices in listed.items() if key[0] != "deepseek-acme-lite"} == {
+        ("fireworks-wildcard", "acme-lite"): (None, None),
+        ("fireworks-wildcard", "acme-max"): (None, None),
+        ("fireworks-wildcard", "acme-pro"): (None, None),
+        ("openai-wildcard", "my-openai/gpt-acme-lite"): (None, None),
+    }
     calls: Final = {
         model_name: router.completion(
             model=model_name, messages=[{"role": "user", "content": "hi"}], mock_response="hello there"