diff --git a/tests/model_executor/test_routed_experts_capture.py b/tests/model_executor/test_routed_experts_capture.py
index 45bf4bcac6a8..bc28db80407f 100644
--- a/tests/model_executor/test_routed_experts_capture.py
+++ b/tests/model_executor/test_routed_experts_capture.py
@@ -15,7 +15,7 @@
 class DummyRouter(BaseRouter):
     @property
     def routing_method_type(self) -> RoutingMethodType:
-        return RoutingMethodType.FUSED_TOPK
+        return RoutingMethodType.TopK
 
     def _compute_routing(self, hidden_states, router_logits, indices_type):
         topk_ids = torch.tensor([[1, 2], [3, 4]], dtype=torch.int64)
@@ -158,3 +158,26 @@ def capture(self, layer_id, topk_ids):
     assert callable(dummy_module.router.capture_fn)
     dummy_module.router.capture_fn(torch.tensor([[9, 10]]))
     assert len(capturer.calls) == 1
+
+
+@pytest.mark.parametrize(
+    "scoring_func,top_k,renormalize,expected",
+    [
+        ("sigmoid", 1, False, RoutingMethodType.Llama4),
+        ("sigmoid", 2, False, RoutingMethodType.DeepSeekV3),
+        ("softmax", 2, False, RoutingMethodType.Default),
+        ("softmax", 2, True, RoutingMethodType.Renormalize),
+    ],
+)
+def test_routing_method_type_from_topk_mapping(
+    scoring_func,
+    top_k,
+    renormalize,
+    expected,
+):
+    assert RoutingMethodType.from_topk(scoring_func, top_k, renormalize) == expected
+
+
+def test_routing_method_type_from_topk_invalid_scoring_func():
+    with pytest.raises(ValueError, match="Unsupported scoring function"):
+        RoutingMethodType.from_topk("none", 1, False)
diff --git a/vllm/model_executor/layers/fused_moe/config.py b/vllm/model_executor/layers/fused_moe/config.py
index 6650367da035..5bd06e10a421 100644
--- a/vllm/model_executor/layers/fused_moe/config.py
+++ b/vllm/model_executor/layers/fused_moe/config.py
@@ -123,6 +123,24 @@ class RoutingMethodType(IntEnum):
     # Unspecified
     Unspecified = 8.0
 
+    @staticmethod
+    def from_topk(
+        scoring_func: str,
+        top_k: int,
+        renormalize: bool,
+    ) -> "RoutingMethodType":
+        if scoring_func == "sigmoid":
+            return (
+                RoutingMethodType.Llama4 if top_k == 1 else RoutingMethodType.DeepSeekV3
+            )
+        if scoring_func == "softmax":
+            return (
+                RoutingMethodType.Renormalize
+                if renormalize
+                else RoutingMethodType.Default
+            )
+        raise ValueError(f"Unsupported scoring function: {scoring_func}")
+
 
 @dataclass
 class FusedMoEQuantDesc:
diff --git a/vllm/model_executor/layers/fused_moe/flashinfer_trtllm_moe.py b/vllm/model_executor/layers/fused_moe/flashinfer_trtllm_moe.py
index 0182cfc195f9..0d59cee0ae5e 100644
--- a/vllm/model_executor/layers/fused_moe/flashinfer_trtllm_moe.py
+++ b/vllm/model_executor/layers/fused_moe/flashinfer_trtllm_moe.py
@@ -63,6 +63,7 @@ def _supports_routing_method(
     """Monolithic kernels need to express router support."""
     if (weight_key, activation_key) == (kFp8Static128BlockSym, kFp8Dynamic128Sym):
         # NOTE(rob): potentially allow others here. This is a conservative list.
+        # Default routing is not implemented in FlashInfer TRTLLM.
         return routing_method in [
             RoutingMethodType.DeepSeekV3,
             RoutingMethodType.Renormalize,
@@ -85,7 +86,6 @@ def _supports_routing_method_bf16(
     routing_method: RoutingMethodType,
 ) -> bool:
     return routing_method in [
-        RoutingMethodType.Default,
         RoutingMethodType.Renormalize,
         RoutingMethodType.DeepSeekV3,
         RoutingMethodType.Llama4,
diff --git a/vllm/model_executor/layers/fused_moe/router/fused_topk_bias_router.py b/vllm/model_executor/layers/fused_moe/router/fused_topk_bias_router.py
index 7c230686fa6e..860318ffbd65 100644
--- a/vllm/model_executor/layers/fused_moe/router/fused_topk_bias_router.py
+++ b/vllm/model_executor/layers/fused_moe/router/fused_topk_bias_router.py
@@ -158,10 +158,8 @@ def __init__(
 
     @property
     def routing_method_type(self) -> RoutingMethodType:
-        return (
-            RoutingMethodType.Renormalize
-            if not self.renormalize
-            else RoutingMethodType.RenormalizeNaive
+        return RoutingMethodType.from_topk(
+            self.scoring_func, self.top_k, self.renormalize
         )
 
     def _compute_routing(
diff --git a/vllm/model_executor/layers/fused_moe/router/fused_topk_router.py b/vllm/model_executor/layers/fused_moe/router/fused_topk_router.py
index cec9240efecd..ba47736f5038 100644
--- a/vllm/model_executor/layers/fused_moe/router/fused_topk_router.py
+++ b/vllm/model_executor/layers/fused_moe/router/fused_topk_router.py
@@ -135,10 +135,8 @@ def __init__(
 
     @property
     def routing_method_type(self) -> RoutingMethodType:
-        return (
-            RoutingMethodType.Renormalize
-            if not self.renormalize
-            else RoutingMethodType.RenormalizeNaive
+        return RoutingMethodType.from_topk(
+            self.scoring_func, self.top_k, self.renormalize
         )
 
     def _compute_routing(
diff --git a/vllm/model_executor/layers/quantization/utils/flashinfer_fp4_moe.py b/vllm/model_executor/layers/quantization/utils/flashinfer_fp4_moe.py
index 4783ca5e0e05..adaa6360bce7 100644
--- a/vllm/model_executor/layers/quantization/utils/flashinfer_fp4_moe.py
+++ b/vllm/model_executor/layers/quantization/utils/flashinfer_fp4_moe.py
@@ -351,7 +351,7 @@ def flashinfer_trtllm_fp4_moe(
         local_expert_offset=layer.ep_rank * layer.local_num_experts,
         local_num_experts=layer.local_num_experts,
         routed_scaling_factor=None,
-        routing_method_type=routing_method_type,
+        routing_method_type=1,
         do_finalize=True,
     )[0]