diff --git a/CHANGELOG.md b/CHANGELOG.md
index b1341a52ab..fa87cbc131 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -15,6 +15,9 @@ Increment the:
 
 ## [Unreleased]
 
+* [TEST] Remove workaround for metrics cardinality limit test
+  [#3663](https://github.com/open-telemetry/opentelemetry-cpp/pull/3663)
+
 ## [1.23 2025-09-25]
 
 * [CodeHealth] Fix clang-tidy warnings part 6
diff --git a/sdk/test/metrics/sum_aggregation_test.cc b/sdk/test/metrics/sum_aggregation_test.cc
index 02421a9e26..5c29fe43e8 100644
--- a/sdk/test/metrics/sum_aggregation_test.cc
+++ b/sdk/test/metrics/sum_aggregation_test.cc
@@ -194,11 +194,7 @@ TEST(HistogramToSumFilterAttributesWithCardinalityLimit, Double)
       {
         for (const MetricData &md : smd.metric_data_)
         {
-          // Something weird about attributes hashmap. If cardinality is setup to n, it emits n-1
-          // including overflow. Just making the logic generic here to succeed for n or n-1 total
-          // cardinality.
-          EXPECT_GE(cardinality_limit, md.point_data_attr_.size());
-          EXPECT_LT(cardinality_limit / 2, md.point_data_attr_.size());
+          EXPECT_EQ(cardinality_limit, md.point_data_attr_.size());
           for (size_t i = 0; i < md.point_data_attr_.size(); i++)
           {
             EXPECT_EQ(1, md.point_data_attr_[i].attributes.size());
@@ -380,8 +376,7 @@ TEST(CounterToSumFilterAttributesWithCardinalityLimit, Double)
           // resulting in a total of cardinality_limit sets. This test checks that the number of
           // emitted attribute sets is within the expected range, accounting for the overflow
           // behavior.
-          EXPECT_GE(cardinality_limit, md.point_data_attr_.size());
-          EXPECT_LT(cardinality_limit / 2, md.point_data_attr_.size());
+          EXPECT_EQ(cardinality_limit, md.point_data_attr_.size());
           for (size_t i = 0; i < md.point_data_attr_.size(); i++)
           {
             EXPECT_EQ(1, md.point_data_attr_[i].attributes.size());