diff --git a/python/sglang/srt/model_executor/model_runner.py b/python/sglang/srt/model_executor/model_runner.py
index 2e4497cd5..0660c9687 100644
--- a/python/sglang/srt/model_executor/model_runner.py
+++ b/python/sglang/srt/model_executor/model_runner.py
@@ -2365,8 +2365,9 @@ class ModelRunner:
         backend_str = self.server_args.moe_runner_backend
         if backend_str not in [
             "flashinfer_trtllm",
-            "flashinfer_cutlass",
             "flashinfer_mxfp4",
+            # TODO: flashinfer_cutlass will cause some flashinfer compilation errors. To be fixed.
+            # "flashinfer_cutlass",
         ]:
             return False
 
diff --git a/test/nightly/test_deepseek_v3_fp4_cutlass_moe.py b/test/nightly/test_deepseek_v3_fp4_cutlass_moe.py
deleted file mode 100644
index c3a509efa..000000000
--- a/test/nightly/test_deepseek_v3_fp4_cutlass_moe.py
+++ /dev/null
@@ -1,75 +0,0 @@
-import unittest
-from types import SimpleNamespace
-
-from sglang.srt.utils import kill_process_tree
-from sglang.test.ci.ci_register import register_cuda_ci
-from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
-from sglang.test.test_utils import (
-    DEFAULT_URL_FOR_TEST,
-    CustomTestCase,
-    is_in_ci,
-    popen_launch_server,
-    write_github_step_summary,
-)
-
-register_cuda_ci(est_time=900, suite="nightly-4-gpu-b200", nightly=True)
-
-FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3-0324-FP4"
-SERVER_LAUNCH_TIMEOUT = 1000
-
-
-class TestDeepseekV3FP4CutlassMoE(CustomTestCase):
-    @classmethod
-    def setUpClass(cls):
-        cls.model = FULL_DEEPSEEK_V3_FP4_MODEL_PATH
-        cls.base_url = DEFAULT_URL_FOR_TEST
-        other_args = [
-            "--tp",
-            "4",
-            "--ep",
-            "4",
-            "--attention-backend",
-            "trtllm_mla",
-            "--moe-runner-backend",
-            "flashinfer_cutlass",
-            "--quantization",
-            "modelopt_fp4",
-            "--model-loader-extra-config",
-            '{"enable_multithread_load": true}',
-        ]
-        cls.process = popen_launch_server(
-            cls.model,
-            cls.base_url,
-            timeout=SERVER_LAUNCH_TIMEOUT,
-            other_args=other_args,
-        )
-
-    @classmethod
-    def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
-
-    def test_a_gsm8k(
-        self,
-    ):  # Append an "a" to make this test run first (alphabetically) to warm up the server
-        args = SimpleNamespace(
-            num_shots=8,
-            data_path=None,
-            num_questions=1319,
-            parallel=1319,
-            max_new_tokens=512,
-            host="http://127.0.0.1",
-            port=int(self.base_url.split(":")[-1]),
-        )
-        metrics = run_eval_few_shot_gsm8k(args)
-        print(f"{metrics=}")
-
-        if is_in_ci():
-            write_github_step_summary(
-                f"### test_gsm8k (deepseek-v3-fp4-cutlass-moe)\n"
-                f'{metrics["accuracy"]=:.3f}\n'
-            )
-            self.assertGreater(metrics["accuracy"], 0.935)
-
-
-if __name__ == "__main__":
-    unittest.main()
diff --git a/test/run_suite_nightly.py b/test/run_suite_nightly.py
index 676d598cd..6e6c701b0 100644
--- a/test/run_suite_nightly.py
+++ b/test/run_suite_nightly.py
@@ -23,7 +23,6 @@ suites = {
         TestFile("test_flashinfer_trtllm_gen_moe_backend.py", 300),
         TestFile("test_gpt_oss_4gpu_perf.py", 600),
         TestFile("test_flashinfer_trtllm_gen_attn_backend.py", 300),
-        TestFile("test_deepseek_v3_fp4_cutlass_moe.py", 900),
         TestFile("test_fp4_moe.py", 300),
         TestFile("test_qwen3_fp4_trtllm_gen_moe.py", 300),
         TestFile("test_eagle_infer_beta_dp_attention_large.py", 600),
diff --git a/test/srt/test_deepseek_v3_fp4_4gpu.py b/test/srt/test_deepseek_v3_fp4_4gpu.py
index e173cb6f1..faf291fa5 100644
--- a/test/srt/test_deepseek_v3_fp4_4gpu.py
+++ b/test/srt/test_deepseek_v3_fp4_4gpu.py
@@ -1,3 +1,4 @@
+import os
 import unittest
 from types import SimpleNamespace
 
@@ -149,5 +150,62 @@ class TestDeepseekV3FP4PiecewiseCudaGraph(CustomTestCase):
         self.assertGreater(speed, 120)
 
 
+class TestDeepseekV3FP4CutlassMoE(CustomTestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = FULL_DEEPSEEK_V3_FP4_MODEL_PATH
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        other_args = [
+            "--tp",
+            "4",
+            "--ep",
+            "4",
+            "--attention-backend",
+            "trtllm_mla",
+            "--moe-runner-backend",
+            "flashinfer_cutlass",
+            "--quantization",
+            "modelopt_fp4",
+            "--model-loader-extra-config",
+            '{"enable_multithread_load": true}',
+        ]
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=SERVER_LAUNCH_TIMEOUT,
+            other_args=other_args,
+            env={
+                **os.environ,
+                "SGLANG_MOE_NVFP4_DISPATCH": "1",  # Enable nvfp4 all gather
+            },
+        )
+
+    @classmethod
+    def tearDownClass(cls):
+        kill_process_tree(cls.process.pid)
+
+    def test_a_gsm8k(
+        self,
+    ):  # Append an "a" to make this test run first (alphabetically) to warm up the server
+        args = SimpleNamespace(
+            num_shots=8,
+            data_path=None,
+            num_questions=1319,
+            parallel=1319,
+            max_new_tokens=512,
+            host="http://127.0.0.1",
+            port=int(self.base_url.split(":")[-1]),
+        )
+        metrics = run_eval_few_shot_gsm8k(args)
+        print(f"{metrics=}")
+
+        if is_in_ci():
+            write_github_step_summary(
+                f"### test_gsm8k (deepseek-v3-fp4-cutlass-moe)\n"
+                f'{metrics["accuracy"]=:.3f}\n'
+            )
+            self.assertGreater(metrics["accuracy"], 0.935)
+
+
 if __name__ == "__main__":
     unittest.main()