diff --git a/backends/webgpu/test/op_tests/cases.py b/backends/webgpu/test/op_tests/cases.py
index 62ecdaec3d2..205ece53601 100644
--- a/backends/webgpu/test/op_tests/cases.py
+++ b/backends/webgpu/test/op_tests/cases.py
@@ -669,3 +669,29 @@ def _max_pool2d_suite() -> WebGPUTestSuite:
         atol=1e-4,
         rtol=1e-3,
     )
+from executorch.backends.webgpu.test.ops.test_relu import (
+    _det_input as _relu_det_input,
+    ReluModule,
+)
+
+
+@register_op_test("relu")
+def _relu_suite() -> WebGPUTestSuite:
+    # SAM2/SAM3 mask-decoder MLP path; tiny + a decoder_mlp-shaped case.
+    return WebGPUTestSuite(
+        module_factory=lambda: ReluModule(),
+        cases=[
+            Case(
+                name="tiny",
+                construct={},
+                inputs=(InputSpec(shape=(1, 4, 8), gen=_relu_det_input),),
+            ),
+            Case(
+                name="decoder_mlp",
+                construct={},
+                inputs=(InputSpec(shape=(1, 576, 768), gen=_relu_det_input),),
+            ),
+        ],
+        atol=1e-4,
+        rtol=1e-3,
+    )
diff --git a/backends/webgpu/test/ops/test_relu.py b/backends/webgpu/test/ops/test_relu.py
new file mode 100644
index 00000000000..f8303fcde58
--- /dev/null
+++ b/backends/webgpu/test/ops/test_relu.py
@@ -0,0 +1,27 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+"""`aten.relu.default` module for the WebGPU op-test framework.
+
+ReLU is on the SAM2/SAM3 mask-decoder MLP path.
+"""
+
+import torch
+
+
+class ReluModule(torch.nn.Module):
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return torch.relu(x)
+
+
+def _det_input(shape) -> torch.Tensor:
+    # ((i % 23) - 11) / 16: exact in fp32, spans negatives through positives
+    # so the clamp is exercised, not just an identity pass-through.
+    n = 1
+    for s in shape:
+        n *= s
+    idx = torch.arange(n, dtype=torch.int64)
+    return (((idx % 23) - 11).to(torch.float32) / 16.0).reshape(shape)