diff --git a/tests/test_maximum.py b/tests/test_maximum.py
new file mode 100644
index 00000000..9ed803fd
--- /dev/null
+++ b/tests/test_maximum.py
@@ -0,0 +1,70 @@
+import pytest
+import torch
+
+import infini.ops
+
+from tests.utils import Payload, empty_strided, get_stream, randn_strided
+
+
+_SHAPE_CASES = (
+    ((13, 4), None, None, None),
+    ((13, 4), (10, 1), (10, 1), (10, 1)),
+    ((13, 4, 4), None, None, None),
+    ((13, 4, 4), (20, 4, 1), (20, 4, 1), (20, 4, 1)),
+    ((16, 5632), None, None, None),
+    ((4, 4, 5632), None, None, None),
+)
+
+_FLOAT_DTYPE_CASES = (
+    (torch.float32, 0.0, 0.0),
+    (torch.float16, 0.0, 0.0),
+    (torch.bfloat16, 0.0, 0.0),
+)
+
+
+@pytest.mark.auto_act_and_assert
+@pytest.mark.parametrize(
+    "shape, input_strides, other_strides, out_strides", _SHAPE_CASES
+)
+@pytest.mark.parametrize(("dtype", "rtol", "atol"), _FLOAT_DTYPE_CASES)
+def test_maximum(
+    shape,
+    input_strides,
+    other_strides,
+    out_strides,
+    dtype,
+    device,
+    implementation_index,
+    rtol,
+    atol,
+):
+    input = randn_strided(shape, input_strides, dtype=dtype, device=device)
+    other = randn_strided(shape, other_strides, dtype=dtype, device=device)
+    out = empty_strided(shape, out_strides, dtype=dtype, device=device)
+
+    return Payload(
+        lambda input, other, out: _maximum(input, other, out, implementation_index),
+        _torch_maximum,
+        (input, other, out),
+        {},
+        rtol=rtol,
+        atol=atol,
+    )
+
+
+def _maximum(input, other, out, implementation_index):
+    infini.ops.maximum(
+        input,
+        other,
+        out,
+        stream=get_stream(input.device),
+        implementation_index=implementation_index,
+    )
+
+    return out
+
+
+def _torch_maximum(input, other, out):
+    out.copy_(torch.maximum(input, other))
+
+    return out