diff --git a/sglang_omni/platforms/musa.py b/sglang_omni/platforms/musa.py
index 4d2cf77e1d1..c872442d43d 100644
--- a/sglang_omni/platforms/musa.py
+++ b/sglang_omni/platforms/musa.py
@@ -33,9 +33,6 @@ def get_joint_rope_inplace_kernel(self) -> None:
         # its own implementation and validation of this capability.
         return None
 
-    def enable_tts_predictor_graph(self) -> bool:
-        return False
-
     def apply_model_worker_backend_policy(
         self,
         server_args: ServerArgs,
diff --git a/tests/unit_test/test_platforms.py b/tests/unit_test/test_platforms.py
index 730642d5293..ad61ed0bae2 100644
--- a/tests/unit_test/test_platforms.py
+++ b/tests/unit_test/test_platforms.py
@@ -289,6 +289,12 @@ def test_xpu_captures_the_qwen3_tts_code_predictor() -> None:
     assert CPUOmniPlatform().enable_tts_predictor_graph() is True
 
 
+def test_musa_captures_the_qwen3_tts_code_predictor() -> None:
+    from sglang_omni.platforms.musa import MUSAOmniPlatform
+
+    assert MUSAOmniPlatform().enable_tts_predictor_graph() is True
+
+
 def test_each_platform_names_the_graph_backend_its_hardware_uses() -> None:
     """The accelerators that capture name a backend; the rest answer None.