diff --git a/paimon-common/src/main/java/org/apache/paimon/data/columnar/heap/HeapBytesVector.java b/paimon-common/src/main/java/org/apache/paimon/data/columnar/heap/HeapBytesVector.java
index 3c57c9bd8003..cbea2f364088 100644
--- a/paimon-common/src/main/java/org/apache/paimon/data/columnar/heap/HeapBytesVector.java
+++ b/paimon-common/src/main/java/org/apache/paimon/data/columnar/heap/HeapBytesVector.java
@@ -74,8 +74,6 @@ public void reset() {
         }
 
         // We don't reset buffer to avoid unnecessary copy.
-        Arrays.fill(buffer, (byte) 0);
-
         this.bytesAppended = 0;
     }
 
diff --git a/paimon-common/src/test/java/org/apache/paimon/data/columnar/heap/HeapBytesVectorReserveBytesTest.java b/paimon-common/src/test/java/org/apache/paimon/data/columnar/heap/HeapBytesVectorReserveBytesTest.java
index b02c745968f5..ac39cded4d7d 100644
--- a/paimon-common/src/test/java/org/apache/paimon/data/columnar/heap/HeapBytesVectorReserveBytesTest.java
+++ b/paimon-common/src/test/java/org/apache/paimon/data/columnar/heap/HeapBytesVectorReserveBytesTest.java
@@ -179,4 +179,20 @@ void testResetClearsBytesAppended() {
         assertThat(bytes.len).isEqualTo(2);
         assertThat(vector.buffer[0]).isEqualTo((byte) 10);
     }
+
+    @Test
+    void testResetDoesNotWipeBuffer() {
+        HeapBytesVector vector = new HeapBytesVector(4);
+        byte[] data = new byte[] {1, 2, 3};
+        vector.putByteArray(0, data, 0, data.length);
+
+        vector.reset();
+
+        // reset() deliberately leaves the data buffer untouched: wiping it costs
+        // O(buffer size) per batch in the vectorized reader hot path, and reads are
+        // always bounded by the start/length offsets, which reset() does clear
+        assertThat(vector.buffer[0]).isEqualTo((byte) 1);
+        assertThat(vector.start[0]).isZero();
+        assertThat(vector.length[0]).isZero();
+    }
 }
diff --git a/paimon-format/src/main/java/org/apache/paimon/format/parquet/reader/VectorizedDeltaByteArrayReader.java b/paimon-format/src/main/java/org/apache/paimon/format/parquet/reader/VectorizedDeltaByteArrayReader.java
index 05ca1bacbbc8..41a7cdedf066 100644
--- a/paimon-format/src/main/java/org/apache/paimon/format/parquet/reader/VectorizedDeltaByteArrayReader.java
+++ b/paimon-format/src/main/java/org/apache/paimon/format/parquet/reader/VectorizedDeltaByteArrayReader.java
@@ -93,11 +93,11 @@ private void readValues(int total, WritableBytesVector c, int rowId) {
 
             c.putByteArray(rowId + i, bytes, offset, length);
             // Keep the value we just assembled rather than a view over the output vector's
-            // buffer. The record reader resets that vector between batches, and reset() zeroes
-            // the buffer, so a view into it would hand NUL bytes to the next value's prefix
-            // whenever a page spans more than one batch. skipBinary avoids the same hazard by
-            // alternating two vectors; here the array is already a fresh copy, so wrapping it
-            // costs nothing.
+            // buffer. The record reader reuses that vector across batches, rewriting it
+            // from offset 0, so a view into it would hand the next batch's bytes to the
+            // next value's prefix whenever a page spans more than one batch. skipBinary
+            // avoids the same hazard by alternating two vectors; here the array is
+            // already a fresh copy, so wrapping it costs nothing.
             previous = ByteBuffer.wrap(bytes);
             currentRow++;
         }
diff --git a/paimon-format/src/test/java/org/apache/paimon/format/parquet/reader/DeltaByteArrayEncodingTest.java b/paimon-format/src/test/java/org/apache/paimon/format/parquet/reader/DeltaByteArrayEncodingTest.java
index f1330f2f9cd0..f21e6db87016 100644
--- a/paimon-format/src/test/java/org/apache/paimon/format/parquet/reader/DeltaByteArrayEncodingTest.java
+++ b/paimon-format/src/test/java/org/apache/paimon/format/parquet/reader/DeltaByteArrayEncodingTest.java
@@ -102,6 +102,23 @@ public void testNegativeSize() {
                         Integer.MAX_VALUE - 1));
     }
 
+    /**
+     * skipBinary alternates two vectors and leaves {@code previous} pointing into the buffer of
+     * whichever one it wrote last, so after an odd number of skipped values the next skip call
+     * starts by resetting that very vector. The prefix of the following value is copied out of it.
+     */
+    @Test
+    public void skippingAnOddNumberOfValuesKeepsThePrefix() throws Exception {
+        String[] vals = new String[] {"aaaa", "aaab", "aaac", "aaad"};
+        Utils.writeData(writer, vals);
+        reader.initFromPage(vals.length, writer.getBytes().toInputStream());
+
+        reader.skipBinary(1);
+        reader.skipBinary(1);
+
+        assertArrayEquals(vals[2].getBytes(), reader.readBinary(0).getBytes());
+    }
+
     /**
      * The record reader resets the vector between batches, so a page that spans two batches has to
      * survive that reset. Every other case here reads a whole page into one vector without a reset,