fix(proxy): expose configured model mode

litellm/proxy/utils.py

@@ -7531,6 +7531,9 @@ def create_model_info_response(
             max_input_tokens = configured_input
         if configured_output is not None:
             max_output_tokens = configured_output
+        configured_mode: Final = llm_router.get_configured_mode(model_id)
+        if isinstance(configured_mode, str):
+            base["mode"] = configured_mode
 
     if max_input_tokens is not None:
         base["max_input_tokens"] = max_input_tokens

litellm/router.py

@@ -9981,6 +9981,17 @@ class Router:
             coerce_token_limit(model_info.get("max_output_tokens")),
         )
 
+    def get_configured_mode(self, model_name: str) -> "str | None":
+        """Return the mode explicitly configured for a concrete deployment."""
+        deployment: Final = self.get_deployment_by_model_group_name(model_group_name=model_name)
+        if deployment is None:
+            return None
+
+        mode: Final = deployment.model_info.get("mode")
+        if isinstance(mode, str) and mode.strip():
+            return mode
+        return None
+
     def get_configured_display_name(self, model_name: str) -> "str | None":
         """
         Return the display_name explicitly configured in a concrete deployment's

litellm/types/proxy/model_listing.py

@@ -11,8 +11,8 @@ class ModelInfoMetadata(TypedDict):
 
 class ModelInfoResponse(TypedDict):
     """OpenAI-compatible model object. `mode`, `max_input_tokens`, and
-    `max_output_tokens` are attached when the cost map knows them; `metadata`
-    is present only when the endpoint is called with include_metadata=true.
+    `max_output_tokens` are attached when the cost map or deployment config
+    knows them; `metadata` is present only with include_metadata=true.
     """
 
     id: str

tests/test_litellm/proxy/test_proxy_utils.py

@@ -942,6 +942,47 @@ def test_create_model_info_response_uses_deployment_limits_when_not_in_cost_map(
     assert response["max_output_tokens"] == 8000
 
 
+def test_create_model_info_response_uses_deployment_mode_for_auto_router():
+    router = litellm.Router(
+        model_list=[
+            {
+                "model_name": "claude-sonnet",
+                "litellm_params": {"model": "openai/gpt-4o-mini", "api_key": "test-key"},
+            },
+            {
+                "model_name": "claude-auto",
+                "litellm_params": {
+                    "model": "auto_router/complexity_router",
+                    "complexity_router_config": {
+                        "tiers": {
+                            "SIMPLE": "claude-sonnet",
+                            "MEDIUM": "claude-sonnet",
+                            "COMPLEX": "claude-sonnet",
+                        }
+                    },
+                    "complexity_router_default_model": "claude-sonnet",
+                },
+                "model_info": {
+                    "mode": "chat",
+                    "max_input_tokens": 1_000_000,
+                    "max_output_tokens": 128_000,
+                },
+            },
+        ]
+    )
+
+    response = create_model_info_response(
+        model_id="claude-auto",
+        provider="openai",
+        llm_router=router,
+        get_model_info=_raise_unmapped,
+    )
+
+    assert response["mode"] == "chat"
+    assert response["max_input_tokens"] == 1_000_000
+    assert response["max_output_tokens"] == 128_000
+
+
 def test_create_model_info_response_deployment_limits_override_cost_map():
     router = MagicMock()
     router.get_configured_token_limits.return_value = (200000, None)