fix(router): pass container create/list through when model names no deployment

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/router.py

@@ -6590,8 +6590,9 @@ class Router:
         upstream value, and route through ``_ageneric_api_call_with_fallbacks`` so
         deployment credentials (e.g. regional ``api_base`` for Azure) match
         :meth:`_init_responses_api_endpoints`. Create/list calls carry no container ID, so
-        they route through the deployment named by ``model`` when the caller passes one.
-        Otherwise call the handler directly with global provider credentials.
+        they route through the deployment named by ``model`` when the caller passes one,
+        falling back to the direct call when no deployment matches. Otherwise call the
+        handler directly with global provider credentials.
         """
         if custom_llm_provider and "custom_llm_provider" not in kwargs:
             kwargs["custom_llm_provider"] = custom_llm_provider
@@ -6627,6 +6628,7 @@ class Router:
         if isinstance(requested_model, str) and requested_model.strip():
             return await self._ageneric_api_call_with_fallbacks(
                 original_function=original_function,
+                passthrough_on_no_deployment=True,
                 **kwargs,
             )
 

tests/router_unit_tests/test_router_endpoints.py

@@ -1383,6 +1383,39 @@ async def test_init_containers_api_endpoints_create_without_model_calls_directly
     mock_original_function.assert_called_once_with(custom_llm_provider="openai", name="Test Container", model=None)
 
 
+@pytest.mark.asyncio
+async def test_init_containers_api_endpoints_create_with_unknown_model_passes_through(monkeypatch):
+    """
+    A ``model`` that names no configured deployment must not turn into a 400. The call
+    falls through to the handler with the caller's model and no injected deployment
+    credentials, matching the behaviour before model-based routing existed.
+    """
+    monkeypatch.delenv("OPENAI_API_KEY", raising=False)
+    router = Router(
+        model_list=[
+            {
+                "model_name": "gpt-5.4",
+                "litellm_params": {"model": "openai/gpt-5.4", "api_key": "sk-model-list-key"},
+            }
+        ]
+    )
+    mock_original_function = AsyncMock(return_value={"id": "cntr_test"})
+
+    await router._init_containers_api_endpoints(
+        original_function=mock_original_function,
+        custom_llm_provider="openai",
+        name="Test Container",
+        model="does-not-exist",
+    )
+
+    mock_original_function.assert_called_once()
+    call_kw = mock_original_function.call_args.kwargs
+    assert call_kw["model"] == "does-not-exist"
+    assert call_kw["name"] == "Test Container"
+    assert "api_key" not in call_kw
+    assert "api_base" not in call_kw
+
+
 def test_router_model_group_encrypted_content_affinity_callback_registration():
     from litellm.router_utils.pre_call_checks.deployment_affinity_check import (
         DeploymentAffinityCheck,

tests/test_litellm/containers/test_container_api.py

@@ -172,7 +172,7 @@ class TestContainerAPI:
         async def _resolve_upstream():
             return upstream_response
 
-        with patch.object(  # test-quality-ok: create_container does not forward a client, so the handler is the only seam
+        with patch.object(  # test-quality-ok: create_container exposes no client seam, only the handler
             base_llm_http_handler,
             "container_create_handler",
             side_effect=lambda **kwargs: _resolve_upstream() if kwargs["_is_async"] else upstream_response,