diff --git a/mediapipe/tasks/c/text/text_classifier/text_classifier_test.cc b/mediapipe/tasks/c/text/text_classifier/text_classifier_test.cc
index 1dd3f69101..af3f30fec6 100644
--- a/mediapipe/tasks/c/text/text_classifier/text_classifier_test.cc
+++ b/mediapipe/tasks/c/text/text_classifier/text_classifier_test.cc
@@ -65,7 +65,7 @@ TEST(TextClassifierTest, SmokeTest) {
   EXPECT_EQ(result.classifications[0].categories_count, 2);
   EXPECT_EQ(std::string{result.classifications[0].categories[0].category_name},
             "positive");
-  EXPECT_NEAR(result.classifications[0].categories[0].score, 0.999371,
+  EXPECT_NEAR(result.classifications[0].categories[0].score, 0.999465,
               kPrecision);
 
   text_classifier_close_result(&result);
diff --git a/mediapipe/tasks/c/text/text_embedder/text_embedder_test.cc b/mediapipe/tasks/c/text/text_embedder/text_embedder_test.cc
index 47622edb74..f67b7ccc45 100644
--- a/mediapipe/tasks/c/text/text_embedder/text_embedder_test.cc
+++ b/mediapipe/tasks/c/text/text_embedder/text_embedder_test.cc
@@ -91,7 +91,7 @@ TEST(TextEmbedderTest, SucceedsWithCosineSimilarity) {
   double similarity;
   text_embedder_cosine_similarity(&result0.embeddings[0],
                                   &result1.embeddings[0], &similarity, nullptr);
-  double expected_similarity = 0.98077;
+  double expected_similarity = 0.97565;
   EXPECT_LE(abs(similarity - expected_similarity), kPrecision);
 
   text_embedder_close_result(&result0);
diff --git a/mediapipe/tasks/cc/text/text_classifier/text_classifier_test.cc b/mediapipe/tasks/cc/text/text_classifier/text_classifier_test.cc
index dfb78c07f0..507acbbb6e 100644
--- a/mediapipe/tasks/cc/text/text_classifier/text_classifier_test.cc
+++ b/mediapipe/tasks/cc/text/text_classifier/text_classifier_test.cc
@@ -151,15 +151,11 @@ TEST_F(TextClassifierTest, TextClassifierWithBert) {
       /*head_name=*/"probability"});
 #else
   negative_expected.classifications.emplace_back(Classifications{
-      /*categories=*/{
-          {/*index=*/0, /*score=*/0.956316, /*category_name=*/"negative"},
-          {/*index=*/1, /*score=*/0.043683, /*category_name=*/"positive"}},
+      /*categories=*/{{0, 0.963325, "negative"}, {1, 0.036674, "positive"}},
       /*head_index=*/0,
       /*head_name=*/"probability"});
   positive_expected.classifications.emplace_back(Classifications{
-      /*categories=*/{
-          {/*index=*/1, /*score=*/0.999945, /*category_name=*/"positive"},
-          {/*index=*/0, /*score=*/0.000056, /*category_name=*/"negative"}},
+      /*categories=*/{{1, 0.9999413, "positive"}, {0, 0.000058, "negative"}},
       /*head_index=*/0,
       /*head_name=*/"probability"});
 #endif  // _WIN32
@@ -255,10 +251,8 @@ TEST_F(TextClassifierTest, BertLongPositive) {
   categories.push_back(
       {/*index=*/0, /*score=*/0.023313, /*category_name=*/"negative"});
 #else
-  categories.push_back(
-      {/*index=*/1, /*score=*/0.985889, /*category_name=*/"positive"});
-  categories.push_back(
-      {/*index=*/0, /*score=*/0.014112, /*category_name=*/"negative"});
+  categories.push_back({1, 0.981590, "positive"});
+  categories.push_back({0, 0.018409, "negative"});
 #endif  // _WIN32
 
   expected.classifications.emplace_back(
diff --git a/mediapipe/tasks/cc/text/text_embedder/text_embedder_test.cc b/mediapipe/tasks/cc/text/text_embedder/text_embedder_test.cc
index 76634a922d..e70eacc9a4 100644
--- a/mediapipe/tasks/cc/text/text_embedder/text_embedder_test.cc
+++ b/mediapipe/tasks/cc/text/text_embedder/text_embedder_test.cc
@@ -82,7 +82,7 @@ TEST_F(EmbedderTest, SucceedsWithMobileBert) {
 #elif defined(__FMA__)
   ASSERT_NEAR(result0.embeddings[0].float_embedding[0], 21.3605f, kEpsilon);
 #else
-  ASSERT_NEAR(result0.embeddings[0].float_embedding[0], 19.9016f, kEpsilon);
+  ASSERT_NEAR(result0.embeddings[0].float_embedding[0], 21.1785f, kEpsilon);
 #endif  // _WIN32
 
   MP_ASSERT_OK_AND_ASSIGN(
@@ -92,7 +92,7 @@ TEST_F(EmbedderTest, SucceedsWithMobileBert) {
 #ifdef __FMA__
   ASSERT_NEAR(result1.embeddings[0].float_embedding[0], 21.254150f, kEpsilon);
 #else
-  ASSERT_NEAR(result1.embeddings[0].float_embedding[0], 22.626251f, kEpsilon);
+  ASSERT_NEAR(result1.embeddings[0].float_embedding[0], 20.322639f, kEpsilon);
 #endif
 
   // Check cosine similarity.