Merge branch 'litellm_internal_staging' into litellm_/eager-rosalind-f7833a

litellm/proxy/_types.py

@@ -517,6 +517,11 @@ class LiteLLMRoutes(enum.Enum):
         "/guardrails/apply_guardrail",
     ]
 
+    model_info_routes = [
+        "/model/info",
+        "/v1/model/info",
+    ]
+
     llm_api_routes = (
         openai_routes
         + anthropic_routes
@@ -527,6 +532,7 @@ class LiteLLMRoutes(enum.Enum):
         + mcp_inference_routes
         + litellm_native_routes
         + agent_routes
+        + model_info_routes
     )
     info_routes = [
         "/key/info",
@@ -653,8 +659,8 @@ class LiteLLMRoutes(enum.Enum):
         "/global/spend/all_tag_names",
     ]
 
-    public_routes = set(
-        [
+    public_routes = frozenset(
+        (
             "/routes",
             "/",
             "/health/liveliness",
@@ -669,7 +675,7 @@ class LiteLLMRoutes(enum.Enum):
             "/public/mcp_hub",
             "/public/skill_hub",
             "/public/litellm_model_cost_map",
-        ]
+        )
     )
 
     # Retained for backwards compatibility with JWT auth configs that reference

litellm/proxy/management_endpoints/team_endpoints.py

@@ -22,7 +22,6 @@ from typing import (
     Mapping,
     Optional,
     Protocol,
-    Sequence,
     Tuple,
     TypeVar,
     Union,

ruff-strict-budget.json

@@ -69,7 +69,7 @@
     "limit": 4
   },
   "C405": {
-    "limit": 23
+    "limit": 22
   },
   "C408": {
     "limit": 14

tests/test_litellm/proxy/auth/test_route_checks.py

@@ -445,6 +445,60 @@ def test_virtual_key_llm_api_routes_allows_mcp_inference_endpoints(route, method
     assert result is True
 
 
+@pytest.mark.parametrize("route", ["/model/info", "/v1/model/info"])
+def test_virtual_key_llm_api_routes_allows_model_info(route):
+    """AI API virtual keys must be able to read model metadata (pricing, mode,
+    max_tokens) for the deployments they can already route to. Both the
+    unversioned and /v1 paths are the same handler, so both must be reachable.
+    """
+
+    valid_token = UserAPIKeyAuth(
+        user_id="test_user",
+        allowed_routes=["llm_api_routes"],
+    )
+
+    result = RouteChecks.is_virtual_key_allowed_to_call_route(
+        route=route,
+        valid_token=valid_token,
+        request=_mock_request("GET"),
+    )
+
+    assert result is True
+
+
+@pytest.mark.parametrize("route", ["/model/info", "/v1/model/info"])
+def test_model_info_not_classified_as_llm_api(route):
+    """Membership in `llm_api_routes` must not promote /model/info to an
+    `is_llm_api_route()`. That predicate gates DISABLE_LLM_API_ENDPOINTS,
+    global/virtual-key budget enforcement, enforce_user_param and the JWT
+    x-litellm-team-id attachment; model metadata is a free read and must stay
+    outside all of them.
+    """
+
+    assert RouteChecks.is_llm_api_route(route=route) is False
+
+
+@pytest.mark.parametrize("route", ["/v2/model/info", "/model_group/info"])
+def test_virtual_key_llm_api_routes_denies_other_model_info_routes(route):
+    """The grant is scoped to the two /model/info paths. The paginated Admin UI
+    listing and the model-group endpoint stay outside it.
+    """
+
+    valid_token = UserAPIKeyAuth(
+        user_id="test_user",
+        allowed_routes=["llm_api_routes"],
+    )
+
+    with pytest.raises(HTTPException) as exc_info:
+        RouteChecks.is_virtual_key_allowed_to_call_route(
+            route=route,
+            valid_token=valid_token,
+            request=_mock_request("GET"),
+        )
+
+    assert exc_info.value.status_code == 403
+
+
 def test_spend_logs_v2_classified_as_management_not_llm_api():
     """Paginated spend logs are a management/spend read route, not an LLM API."""
 

type-discipline-budget.json

@@ -3,7 +3,7 @@
     "limit": 23191
   },
   "LIT002": {
-    "limit": 27276
+    "limit": 27275
   },
   "LIT003": {
     "limit": 292