Merge pull request #42121 from BerriAI/litellm_utils_model_info_lookup

feat(proxy): add GET /utils/model_info to look up cost map info for unregistered models

litellm/proxy/_types.py

@@ -403,6 +403,7 @@ class LiteLLMRoutes(enum.Enum):
         "/v1/models",
         # token counter
         "/utils/token_counter",
+        "/utils/model_info",
         "/utils/transform_request",
         # rerank
         "/rerank",

litellm/proxy/auth/auth_checks.py

@@ -845,6 +845,7 @@ MODEL_DISCOVERY_ROUTES: Final = frozenset(
         "/v1/model/info",
         "/v2/model/info",
         "/model_group/info",
+        "/utils/model_info",
     }
 )
 

litellm/proxy/proxy_server.py

@@ -13476,6 +13476,48 @@ async def supported_openai_params(model: str):
         raise HTTPException(status_code=400, detail={"error": f"Could not map model={model}"})
 
 
+class _ModelInfoLookupResponse(TypedDict):
+    model: ReadOnly[str]
+    custom_llm_provider: ReadOnly[str]
+    model_info: ReadOnly[Mapping[str, object]]
+
+
+@router.get(
+    "/utils/model_info",
+    tags=["llm utils"],  # mutable-ok: FastAPI tags kwarg is list-typed
+    dependencies=[Depends(user_api_key_auth)],  # mutable-ok: FastAPI dependencies kwarg is list-typed
+)
+async def model_info_lookup(model: str, custom_llm_provider: str | None = None):
+    """
+    Returns the model cost map entry (token limits, pricing, supports_* capabilities) for any model
+    in the cost map, whether or not it is registered on this proxy. `model_info` carries every
+    field of the raw cost map entry plus the typed fields `litellm.get_model_info` derives from it
+    (`key`, `supported_openai_params`).
+
+    Example curl:
+    ```
+    curl -X GET --location 'http://localhost:4000/utils/model_info?model=gpt-4o&custom_llm_provider=openai' \
+        --header 'Authorization: Bearer sk-1234'
+    ```
+    """
+    detail: Final = {  # mutable-ok: FastAPI serializes detail as a plain dict
+        "error": f"model={model}, custom_llm_provider={custom_llm_provider} is not in the model cost map"
+    }
+    try:
+        typed_model_info: Final = litellm.get_model_info(model=model, custom_llm_provider=custom_llm_provider)
+    except Exception:
+        raise HTTPException(status_code=404, detail=detail)
+    cost_map_entry: Final = litellm.model_cost.get(typed_model_info["key"])
+    if cost_map_entry is None:
+        raise HTTPException(status_code=404, detail=detail)
+    response: Final[_ModelInfoLookupResponse] = {
+        "model": model,
+        "custom_llm_provider": typed_model_info["litellm_provider"],
+        "model_info": {**typed_model_info, **cost_map_entry},
+    }
+    return response
+
+
 @router.post(
     "/utils/transform_request",
     tags=["llm utils"],

tests/test_litellm/proxy/auth/test_auth_checks.py

@@ -67,6 +67,7 @@ from litellm.constants import (
     REGISTRY_ERROR_NEGATIVE_CACHE_TTL,
     TAG_REGISTRY_MAX_SIZE,
 )
+from litellm.proxy.auth.route_checks import RouteChecks
 from litellm.proxy.common_utils.encrypt_decrypt_utils import decrypt_value_helper
 from litellm.proxy.common_utils.user_api_key_cache import (
     END_USER_RESTRICTED_REGISTRY_OVERFLOW_SENTINEL,
@@ -8889,6 +8890,8 @@ def test_jwt_team_role_reaches_the_gateway_token_endpoint_by_default():
 def test_route_skips_budget_checks_marks_only_spend_free_routes() -> None:
     assert route_skips_budget_checks(route="/v1/models") is True
     assert route_skips_budget_checks(route="/spend/logs") is True
+    assert route_skips_budget_checks(route="/utils/model_info") is True
+    assert RouteChecks.is_llm_api_route(route="/utils/model_info") is True
     assert route_skips_budget_checks(route="/health") is False
     assert route_skips_budget_checks(route="/v1/chat/completions") is False
 

tests/test_litellm/proxy/proxy_server/test_routes_utils.py

@@ -3,6 +3,7 @@
 Pins (PR2):
     - POST /utils/token_counter
     - GET /utils/supported_openai_params
+    - GET /utils/model_info
     - POST /utils/transform_request
 """
 
@@ -231,6 +232,66 @@ def test_supported_openai_params_invalid_model(client, auth_as, monkeypatch):
     assert "Could not map model" in response.text
 
 
+# ---------------------------------------------------------------------------
+# GET /utils/model_info
+# ---------------------------------------------------------------------------
+
+
+@pytest.fixture
+def lookup_fixture_model(monkeypatch):
+    entry = {
+        "litellm_provider": "openai",
+        "mode": "chat",
+        "max_input_tokens": 1234,
+        "max_output_tokens": 56,
+        "input_cost_per_token": 1e-6,
+        "output_cost_per_token": 2e-6,
+        "supports_vision": True,
+        "deprecation_date": "2099-01-01",
+        "supports_lookup_fixture_edit": True,
+    }
+    monkeypatch.setattr(proxy_server, "llm_router", None)
+    monkeypatch.setitem(litellm.model_cost, "lookup-fixture-model", entry)
+    litellm.get_model_info.cache_clear()
+    litellm.utils._cached_get_model_info_helper.cache_clear()
+    yield entry
+    litellm.get_model_info.cache_clear()
+    litellm.utils._cached_get_model_info_helper.cache_clear()
+
+
+def test_model_info_lookup_returns_full_cost_map_entry_for_unregistered_model(client, auth_as, lookup_fixture_model):
+    """Every raw cost map field comes back, including ones outside ``ModelInfoBase`` that ``get_model_info`` drops."""
+    with auth_as():
+        response = client.get(
+            "/utils/model_info", params={"model": "lookup-fixture-model", "custom_llm_provider": "openai"}
+        )
+    assert response.status_code == 200, response.text
+    body = response.json()
+    assert body["model"] == "lookup-fixture-model"
+    assert body["custom_llm_provider"] == "openai"
+    assert body["model_info"]["key"] == "lookup-fixture-model"
+    assert isinstance(body["model_info"]["supported_openai_params"], list)
+    assert {k: body["model_info"][k] for k in lookup_fixture_model} == lookup_fixture_model
+
+
+def test_model_info_lookup_unknown_model_returns_404(client, auth_as, monkeypatch):
+    monkeypatch.setattr(proxy_server, "llm_router", None)
+    with auth_as():
+        response = client.get("/utils/model_info", params={"model": "no-such-model-lit-7476"})
+    assert response.status_code == 404, response.text
+    assert "is not in the model cost map" in response.text
+
+
+def test_model_info_lookup_returns_404_when_typed_info_has_no_cost_map_entry(client, auth_as, monkeypatch):
+    """``get_model_info`` synthesizes info for huggingface fallbacks absent from ``model_cost``;
+    with no raw entry the route must 404 rather than answer 200 with typed fields only."""
+    monkeypatch.setattr(proxy_server, "llm_router", None)
+    with auth_as():
+        response = client.get("/utils/model_info", params={"model": "huggingface/not-in-map-org/not-in-map-model"})
+    assert response.status_code == 404, response.text
+    assert "is not in the model cost map" in response.text
+
+
 # ---------------------------------------------------------------------------
 # POST /utils/transform_request
 # ---------------------------------------------------------------------------

ui/litellm-dashboard/src/lib/http/schema.d.ts

@@ -17448,6 +17448,34 @@ export interface paths {
         patch?: never;
         trace?: never;
     };
+    "/utils/model_info": {
+        parameters: {
+            query?: never;
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        /**
+         * Model Info Lookup
+         * @description Returns the model cost map entry (token limits, pricing, supports_* capabilities) for any model
+         *     in the cost map, whether or not it is registered on this proxy. `model_info` carries every
+         *     field of the raw cost map entry plus the typed fields `litellm.get_model_info` derives from it
+         *     (`key`, `supported_openai_params`).
+         *
+         *     Example curl:
+         *     ```
+         *     curl -X GET --location 'http://localhost:4000/utils/model_info?model=gpt-4o&custom_llm_provider=openai'         --header 'Authorization: Bearer sk-1234'
+         *     ```
+         */
+        get: operations["model_info_lookup_utils_model_info_get"];
+        put?: never;
+        post?: never;
+        delete?: never;
+        options?: never;
+        head?: never;
+        patch?: never;
+        trace?: never;
+    };
     "/utils/supported_openai_params": {
         parameters: {
             query?: never;
@@ -63647,6 +63675,38 @@ export interface operations {
             };
         };
     };
+    model_info_lookup_utils_model_info_get: {
+        parameters: {
+            query: {
+                model: string;
+                custom_llm_provider?: string | null;
+            };
+            header?: never;
+            path?: never;
+            cookie?: never;
+        };
+        requestBody?: never;
+        responses: {
+            /** @description Successful Response */
+            200: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "application/json": unknown;
+                };
+            };
+            /** @description Validation Error */
+            422: {
+                headers: {
+                    [name: string]: unknown;
+                };
+                content: {
+                    "application/json": components["schemas"]["HTTPValidationError"];
+                };
+            };
+        };
+    };
     supported_openai_params_utils_supported_openai_params_get: {
         parameters: {
             query: {