diff --git a/litellm/router.py b/litellm/router.py
index 9cde292657c..8f2397173d3 100644
--- a/litellm/router.py
+++ b/litellm/router.py
@@ -4490,7 +4490,15 @@ class Router:
             self._update_kwargs_with_deployment(deployment=deployment, kwargs=kwargs, function_name=function_name)
 
             data: Final = deployment["litellm_params"].copy()
-            model_name: Final = data["model"]
+            deployment_model_name: Final = data["model"]
+            # Wildcard deployments provide provider credentials but must keep
+            # the concrete model requested by the caller. Passing the wildcard
+            # through prevents model-cost lookups and Responses API bridging.
+            model_name = (
+                model
+                if "*" in deployment_model_name and "*" not in model
+                else deployment_model_name
+            )
             self.total_calls[model_name] += 1
 
             self._add_deployment_model_to_endpoint_for_llm_passthrough_route(
diff --git a/tests/router_unit_tests/test_router_wildcard_model.py b/tests/router_unit_tests/test_router_wildcard_model.py
new file mode 100644
index 00000000000..c643d0f8237
--- /dev/null
+++ b/tests/router_unit_tests/test_router_wildcard_model.py
@@ -0,0 +1,36 @@
+from unittest.mock import AsyncMock
+from unittest.mock import patch
+
+import pytest
+
+from litellm import Router
+
+
+@pytest.mark.asyncio
+async def test_wildcard_deployment_preserves_requested_model() -> None:
+    """Provider wildcard credentials must not replace the concrete request model."""
+    router = Router(model_list=[])
+    deployment = {
+        "model_name": "openai/*",
+        "litellm_params": {"model": "openai/*", "api_key": "sk-test"},
+        "model_info": {"id": "wildcard-deployment"},
+    }
+    original_function = AsyncMock(return_value="response")
+
+    with (
+        patch.object(
+            router,
+            "async_get_available_deployment",
+            new=AsyncMock(return_value=deployment),
+        ),
+        patch.object(router, "async_routing_strategy_pre_call_checks", new=AsyncMock()),
+        patch.object(router, "_get_client", return_value=None),
+    ):
+        result = await router._ageneric_api_call_with_fallbacks_helper(
+            model="openai/gpt-5.3-codex",
+            original_generic_function=original_function,
+            messages=[{"role": "user", "content": "ping"}],
+        )
+
+    assert result == "response"
+    assert original_function.await_args.kwargs["model"] == "openai/gpt-5.3-codex"