diff --git a/stl/inc/bit b/stl/inc/bit
index 1ac47cd6a99..bacde1abff9 100644
--- a/stl/inc/bit
+++ b/stl/inc/bit
@@ -82,8 +82,8 @@ _NODISCARD constexpr _Ty bit_floor(const _Ty _Val) noexcept {
 }
 
 template <class _Ty, enable_if_t<_Is_standard_unsigned_integer<_Ty>, int> = 0>
-_NODISCARD constexpr _Ty bit_width(const _Ty _Val) noexcept {
-    return static_cast<_Ty>(numeric_limits<_Ty>::digits - _STD countl_zero(_Val));
+_NODISCARD constexpr int bit_width(const _Ty _Val) noexcept {
+    return numeric_limits<_Ty>::digits - _STD countl_zero(_Val);
 }
 
 template <class _Ty, enable_if_t<_Is_standard_unsigned_integer<_Ty>, int> = 0>
diff --git a/stl/inc/format b/stl/inc/format
index f189bfa7dc8..bd0dfea89f5 100644
--- a/stl/inc/format
+++ b/stl/inc/format
@@ -2939,7 +2939,7 @@ _NODISCARD _OutputIt _Fmt_write(
         // Add 3 to the bit width so we always round up on the division.
         // Divide that by the amount of bits a hexit represents (log2(16) = log2(2^4) = 4).
         // Add 2 for the 0x prefix.
-        _Width = static_cast<int>(2 + (_STD bit_width(reinterpret_cast<uintptr_t>(_Value)) + 3) / 4);
+        _Width = 2 + (_STD bit_width(reinterpret_cast<uintptr_t>(_Value)) + 3) / 4;
     }
 
     return _Write_aligned(_STD move(_Out), _Width, _Specs, _Fmt_align::_Right,
diff --git a/tests/libcxx/expected_results.txt b/tests/libcxx/expected_results.txt
index 0245f44cfae..e180d4efc43 100644
--- a/tests/libcxx/expected_results.txt
+++ b/tests/libcxx/expected_results.txt
@@ -652,6 +652,9 @@ std/utilities/charconv/charconv.msvc/test.pass.cpp FAIL
 std/ranges/range.adaptors/range.transform/end.pass.cpp FAIL
 std/ranges/range.adaptors/range.transform/iterator/base.pass.cpp FAIL
 
+# libc++ doesn't yet implement LWG-3656 (https://reviews.llvm.org/D120444)
+std/numerics/bit/bit.pow.two/bit_width.pass.cpp FAIL
+
 # MaybePOCCAAllocator doesn't meet the allocator requirements
 std/containers/sequences/vector/vector.cons/assign_copy.pass.cpp FAIL
 
diff --git a/tests/libcxx/skipped_tests.txt b/tests/libcxx/skipped_tests.txt
index 5f0d8f4a26f..86379881df9 100644
--- a/tests/libcxx/skipped_tests.txt
+++ b/tests/libcxx/skipped_tests.txt
@@ -652,6 +652,9 @@ utilities\charconv\charconv.msvc\test.pass.cpp
 ranges\range.adaptors\range.transform\end.pass.cpp
 ranges\range.adaptors\range.transform\iterator\base.pass.cpp
 
+# libc++ doesn't yet implement LWG-3656 (https://reviews.llvm.org/D120444)
+numerics\bit\bit.pow.two\bit_width.pass.cpp
+
 # MaybePOCCAAllocator doesn't meet the allocator requirements
 containers\sequences\vector\vector.cons\assign_copy.pass.cpp
 
diff --git a/tests/std/tests/P0556R3_bit_integral_power_of_two_operations/test.cpp b/tests/std/tests/P0556R3_bit_integral_power_of_two_operations/test.cpp
index 983145ab428..f727684f619 100644
--- a/tests/std/tests/P0556R3_bit_integral_power_of_two_operations/test.cpp
+++ b/tests/std/tests/P0556R3_bit_integral_power_of_two_operations/test.cpp
@@ -3,6 +3,7 @@
 #include <bit>
 #include <cassert>
 #include <limits>
+#include <type_traits>
 #include <utility>
 
 using namespace std;
@@ -76,12 +77,16 @@ constexpr bool test_bit_floor() {
 template <typename T>
 constexpr bool test_bit_width() {
     constexpr int digits = numeric_limits<T>::digits;
-    assert(bit_width(T{0}) == T{0});
+    assert(bit_width(T{0}) == 0);
     assert(bit_width(numeric_limits<T>::max()) == digits);
-    assert(bit_width(T{1}) == T{1});
+    assert(bit_width(T{1}) == 1);
     for (int i = 1; i < digits; ++i) {
-        assert(bit_width(static_cast<T>(T{1} << i)) == static_cast<T>(i + 1));
+        assert(bit_width(static_cast<T>(T{1} << i)) == i + 1);
     }
+
+    // LWG-3656: bit_width returns int
+    static_assert(is_same_v<decltype(bit_width(T{0})), int>);
+
     return true;
 }