diff --git a/src/numpy-stubs/@test/runtime/test_ctype_assumptions.py b/src/numpy-stubs/@test/runtime/test_ctype_assumptions.py
index 63d425c3..df7e5aa5 100644
--- a/src/numpy-stubs/@test/runtime/test_ctype_assumptions.py
+++ b/src/numpy-stubs/@test/runtime/test_ctype_assumptions.py
@@ -1,13 +1,15 @@
 # ruff: noqa: S101, PLR2004
 
 import ctypes as ct
+import platform
 import sys
 
 import pytest
 
 import numpy as np
 
-WIN32 = sys.platform == "win32"
+IS_WIN = sys.platform == "win32"
+IS_X86 = platform.machine().lower() in {"x86_64", "amd64", "i386", "i686"}
 SIZE_P = ct.sizeof(ct.c_ssize_t) * 8
 
 
@@ -19,8 +21,9 @@ def test_maxsize_32_or_64() -> None:
 
 @pytest.mark.parametrize("char", ["q", "Q"])
 def test_longlong_64(char: str) -> None:
-    assert np.dtype(char).itemsize == 8
-    assert np.dtype(char).alignment == 8
+    dt = np.dtype(char)
+    assert dt.itemsize == 8
+    assert dt.alignment == 8
 
 
 @pytest.mark.parametrize(
@@ -29,30 +32,38 @@ def test_longlong_64(char: str) -> None:
         ("byte", "int8"),
         ("short", "int16"),
         ("intc", "int32"),
-        ("long", "int32" if WIN32 else f"int{SIZE_P}"),
         ("intp", f"int{SIZE_P}"),
+        ("long", "int32" if IS_WIN else f"int{SIZE_P}"),
     ],
 )
 def test_alias_integer(name_c: str, name_expect: str) -> None:
     signed_c: type[np.signedinteger] = getattr(np, name_c)
     signed_expect: type[np.signedinteger] = getattr(np, name_expect)
-    assert signed_c is signed_expect
+    assert np.dtype(signed_c) == np.dtype(signed_expect)
 
     unsigned_c: type[np.unsignedinteger] = getattr(np, f"u{name_c}")
     unsigned_expect: type[np.unsignedinteger] = getattr(np, f"u{name_expect}")
-    assert unsigned_c is unsigned_expect
+    assert np.dtype(unsigned_c) == np.dtype(unsigned_expect)
 
 
 @pytest.mark.parametrize(
     ("name_c", "name_expect"),
-    [
-        ("half", "float16"),
-        ("single", "float32"),
-        ("double", "float64"),
-        ("longdouble", "float96" if WIN32 else f"float{SIZE_P + 64}"),
-    ],
+    [("half", "float16"), ("single", "float32"), ("double", "float64")],
 )
-def test_alias_floating(name_c: str, name_expect: str) -> None:
+def test_alias_floating_standard(name_c: str, name_expect: str) -> None:
     floating_c: type[np.floating] = getattr(np, name_c)
     floating_expect: type[np.floating] = getattr(np, name_expect)
-    assert floating_c is floating_expect
+    assert np.dtype(floating_c) == np.dtype(floating_expect)
+
+
+def test_alias_longdouble() -> None:
+    ld_name = np.dtype(np.longdouble).name
+
+    if IS_WIN:
+        assert ld_name in {"float64", "float96"}
+    elif IS_X86:
+        assert ld_name == f"float{SIZE_P + 64}"
+    else:
+        assert ld_name in {"float64", "float128"}
+
+    assert np.dtype(np.longdouble) == np.dtype(getattr(np, ld_name))