diff --git a/NEWS b/NEWS
index 5182713edfc5..b902bbef1086 100644
--- a/NEWS
+++ b/NEWS
@@ -19,6 +19,11 @@ PHP                                                                        NEWS
   . Fixed bug GH-23385 (SplDoublyLinkedList::serialize() use-after-free when
     __serialize() removes an element). (David Carlier)
 
+- JSON:
+  . Added a vectorized (SSE2/NEON) fast path to php_json_escape_string()
+    that bulk-copies runs of bytes which need no escaping, speeding up
+    json_encode() on ASCII-heavy strings.
+
 
 10 Sep 2026, PHP 8.6.0beta3
 
diff --git a/UPGRADING b/UPGRADING
index 3980faa4294d..7e7cfef7f6e5 100644
--- a/UPGRADING
+++ b/UPGRADING
@@ -1083,6 +1083,8 @@ PHP 8.6 UPGRADE NOTES
   . Improve performance of encoding arrays and objects.
   . Improved performance of indentation generation in json_encode()
     when using PHP_JSON_PRETTY_PRINT.
+  . Added an SSE2/NEON fast path for encoding strings that need no
+    escaping, speeding up json_encode() on ASCII-heavy payloads.
 
 - Intl:
   . Improved performance of IntlCalendar::getAvailableLocales() and
diff --git a/ext/json/json_encoder.c b/ext/json/json_encoder.c
index b8ae31040c8b..8080483782ca 100644
--- a/ext/json/json_encoder.c
+++ b/ext/json/json_encoder.c
@@ -27,9 +27,32 @@
 #include "zend_enum.h"
 #include "zend_property_hooks.h"
 #include "zend_lazy_objects.h"
+#include "zend_simd.h"
+#include "zend_bitset.h"
 
 static const char digits[] = "0123456789abcdef";
 
+#ifdef XSSE2
+/* Bytes that need escaping: < 0x20, >= 0x80, and the ASCII specials
+ * " \ / < > & '. Must be kept in sync with the `charmap` bitmap used by
+ * the scalar loop in php_json_escape_string(). _mm_cmplt_epi8() is a
+ * signed compare, so "< 0x20" already covers every byte >= 0x80 (negative
+ * as a signed int8), which is why there's no separate high-bit test. */
+static zend_always_inline __m128i php_json_escape_dirty_mask(__m128i chunk)
+{
+	__m128i dirty = _mm_cmplt_epi8(chunk, _mm_set1_epi8(0x20));
+
+	dirty = _mm_or_si128(dirty, _mm_cmpeq_epi8(chunk, _mm_set1_epi8('"')));
+	dirty = _mm_or_si128(dirty, _mm_cmpeq_epi8(chunk, _mm_set1_epi8('\\')));
+	dirty = _mm_or_si128(dirty, _mm_cmpeq_epi8(chunk, _mm_set1_epi8('/')));
+	dirty = _mm_or_si128(dirty, _mm_cmpeq_epi8(chunk, _mm_set1_epi8('<')));
+	dirty = _mm_or_si128(dirty, _mm_cmpeq_epi8(chunk, _mm_set1_epi8('>')));
+	dirty = _mm_or_si128(dirty, _mm_cmpeq_epi8(chunk, _mm_set1_epi8('&')));
+	dirty = _mm_or_si128(dirty, _mm_cmpeq_epi8(chunk, _mm_set1_epi8('\'')));
+	return dirty;
+}
+#endif
+
 static zend_always_inline bool php_json_check_stack_limit(void)
 {
 #ifdef ZEND_CHECK_STACK_LIMIT
@@ -379,7 +402,8 @@ zend_result php_json_escape_string(
 
 	/* pre-allocate for string length plus 2 quotes */
 	smart_str_alloc(buf, len+2, 0);
-	smart_str_appendc(buf, '"');
+	ZSTR_VAL(buf->s)[ZSTR_LEN(buf->s)] = '"';
+	ZSTR_LEN(buf->s)++;
 
 	pos = 0;
 
@@ -388,6 +412,58 @@ zend_result php_json_escape_string(
 			0xffffffff, 0x500080c4, 0x10000000, 0x00000000,
 			0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff};
 
+#ifdef XSSE2
+#if defined(__aarch64__) || defined(_M_ARM64)
+		while (len >= sizeof(__m128i)) {
+			if (UNEXPECTED(ZEND_BIT_TEST(charmap, (unsigned char) s[pos]))) {
+				break;
+			}
+
+			__m128i chunk = _mm_loadu_si128((const __m128i *)(s + pos));
+			__m128i dirty = php_json_escape_dirty_mask(chunk);
+			uint8x16_t dirty_u8 = vreinterpretq_u8_s8(dirty);
+
+			if (vmaxvq_u8(dirty_u8) == 0) {
+				pos += sizeof(__m128i);
+				len -= sizeof(__m128i);
+				continue;
+			}
+			{
+				static const uint8_t lane_index[16] = {
+					0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15};
+				uint8x16_t masked_idx = vbslq_u8(dirty_u8, vld1q_u8(lane_index), vdupq_n_u8(0xff));
+				size_t clean = vminvq_u8(masked_idx);
+				pos += clean;
+				len -= clean;
+			}
+			break;
+		}
+#else
+		while (len >= sizeof(__m128i)) {
+			if (UNEXPECTED(ZEND_BIT_TEST(charmap, (unsigned char) s[pos]))) {
+				break;
+			}
+
+			__m128i chunk = _mm_loadu_si128((const __m128i *)(s + pos));
+			__m128i dirty = php_json_escape_dirty_mask(chunk);
+
+			int mask = _mm_movemask_epi8(dirty);
+			if (mask != 0) {
+				size_t clean = zend_ulong_ntz((zend_ulong) (unsigned int) mask);
+				pos += clean;
+				len -= clean;
+				break;
+			}
+			pos += sizeof(__m128i);
+			len -= sizeof(__m128i);
+		}
+#endif
+		if (len == 0) {
+			smart_str_appendl(buf, s, pos);
+			break;
+		}
+#endif
+
 		unsigned int us = (unsigned char)s[pos];
 		if (EXPECTED(!ZEND_BIT_TEST(charmap, us))) {
 			pos++;
diff --git a/ext/json/tests/json_encode_sse2_boundary.phpt b/ext/json/tests/json_encode_sse2_boundary.phpt
new file mode 100644
index 000000000000..bb5b396cc7f2
--- /dev/null
+++ b/ext/json/tests/json_encode_sse2_boundary.phpt
@@ -0,0 +1,116 @@
+--TEST--
+json_encode() SSE2 fast-path boundary handling
+--FILE--
+<?php
+/* Regression tests for the SSE2 chunked fast path in
+ * php_json_escape_string() (ext/json/json_encoder.c). The fast path scans
+ * 16-byte chunks and resumes scanning right after each escaped byte, so
+ * exercise lengths and escape/codepoint positions around that boundary. */
+
+function check($label, $actual, $expected) {
+    if ($actual === $expected) {
+        echo "$label: OK\n";
+    } else {
+        echo "$label: FAIL\n";
+        var_dump($expected, $actual);
+    }
+}
+
+// 1. Pure clean ASCII at/around the 16-byte chunk boundary.
+foreach ([15, 16, 17, 31, 32, 33] as $len) {
+    $s = str_repeat('a', $len);
+    check("clean len=$len", json_encode($s), '"' . $s . '"');
+}
+
+// 2. A single escapable byte at each offset around the boundary.
+foreach ([0, 15, 16, 17] as $off) {
+    $len = 24;
+    $s = str_repeat('a', $off) . '"' . str_repeat('a', $len - $off - 1);
+    $expected = '"' . str_repeat('a', $off) . '\\"' . str_repeat('a', $len - $off - 1) . '"';
+    check("quote at offset=$off", json_encode($s), $expected);
+}
+
+// 3. A 3-byte UTF-8 sequence (EUR SIGN, U+20AC) straddling the boundary.
+// Checked both with the default \uXXXX escaping and with
+// JSON_UNESCAPED_UNICODE, which appends the raw UTF-8 bytes instead.
+foreach ([13, 14, 15, 16] as $off) {
+    $len = 20;
+    $s = str_repeat('a', $off) . "\xe2\x82\xac" . str_repeat('a', $len - $off - 3);
+    $tail = str_repeat('a', $len - $off - 3);
+    check("3-byte utf8 at offset=$off", json_encode($s),
+        '"' . str_repeat('a', $off) . "\\u20ac" . $tail . '"');
+    check("3-byte utf8 at offset=$off, UNESCAPED_UNICODE", json_encode($s, JSON_UNESCAPED_UNICODE),
+        '"' . str_repeat('a', $off) . '€' . $tail . '"');
+}
+
+// 4. A 4-byte UTF-8 sequence (surrogate pair, U+1F600) straddling the boundary.
+foreach ([12, 13, 14, 15, 16] as $off) {
+    $len = 20;
+    $s = str_repeat('a', $off) . "\xf0\x9f\x98\x80" . str_repeat('a', $len - $off - 4);
+    $tail = str_repeat('a', $len - $off - 4);
+    check("4-byte utf8 at offset=$off", json_encode($s),
+        '"' . str_repeat('a', $off) . "\\ud83d\\ude00" . $tail . '"');
+    check("4-byte utf8 at offset=$off, UNESCAPED_UNICODE", json_encode($s, JSON_UNESCAPED_UNICODE),
+        '"' . str_repeat('a', $off) . '😀' . $tail . '"');
+}
+
+// 5. Invalid UTF-8 straddling the boundary: confirm the checkpoint/rollback
+// and each INVALID_UTF8_* option still land correctly after the fast path
+// has already appended bytes.
+foreach ([14, 15, 16, 17] as $off) {
+    $len = 20;
+    $s = str_repeat('a', $off) . "\xb0" . str_repeat('a', $len - $off - 1);
+    $tail = str_repeat('a', $len - $off - 1);
+
+    check("invalid utf8 at offset=$off, no flag", json_encode($s), false);
+
+    check("invalid utf8 at offset=$off, IGNORE",
+        json_encode($s, JSON_INVALID_UTF8_IGNORE),
+        '"' . str_repeat('a', $off) . $tail . '"');
+
+    check("invalid utf8 at offset=$off, SUBSTITUTE",
+        json_encode($s, JSON_INVALID_UTF8_SUBSTITUTE),
+        '"' . str_repeat('a', $off) . "\\ufffd" . $tail . '"');
+}
+?>
+--EXPECT--
+clean len=15: OK
+clean len=16: OK
+clean len=17: OK
+clean len=31: OK
+clean len=32: OK
+clean len=33: OK
+quote at offset=0: OK
+quote at offset=15: OK
+quote at offset=16: OK
+quote at offset=17: OK
+3-byte utf8 at offset=13: OK
+3-byte utf8 at offset=13, UNESCAPED_UNICODE: OK
+3-byte utf8 at offset=14: OK
+3-byte utf8 at offset=14, UNESCAPED_UNICODE: OK
+3-byte utf8 at offset=15: OK
+3-byte utf8 at offset=15, UNESCAPED_UNICODE: OK
+3-byte utf8 at offset=16: OK
+3-byte utf8 at offset=16, UNESCAPED_UNICODE: OK
+4-byte utf8 at offset=12: OK
+4-byte utf8 at offset=12, UNESCAPED_UNICODE: OK
+4-byte utf8 at offset=13: OK
+4-byte utf8 at offset=13, UNESCAPED_UNICODE: OK
+4-byte utf8 at offset=14: OK
+4-byte utf8 at offset=14, UNESCAPED_UNICODE: OK
+4-byte utf8 at offset=15: OK
+4-byte utf8 at offset=15, UNESCAPED_UNICODE: OK
+4-byte utf8 at offset=16: OK
+4-byte utf8 at offset=16, UNESCAPED_UNICODE: OK
+invalid utf8 at offset=14, no flag: OK
+invalid utf8 at offset=14, IGNORE: OK
+invalid utf8 at offset=14, SUBSTITUTE: OK
+invalid utf8 at offset=15, no flag: OK
+invalid utf8 at offset=15, IGNORE: OK
+invalid utf8 at offset=15, SUBSTITUTE: OK
+invalid utf8 at offset=16, no flag: OK
+invalid utf8 at offset=16, IGNORE: OK
+invalid utf8 at offset=16, SUBSTITUTE: OK
+invalid utf8 at offset=17, no flag: OK
+invalid utf8 at offset=17, IGNORE: OK
+invalid utf8 at offset=17, SUBSTITUTE: OK
diff --git a/ext/json/tests/json_encode_sse2_fuzz.phpt b/ext/json/tests/json_encode_sse2_fuzz.phpt
new file mode 100644
index 000000000000..2e2b780b17c1
--- /dev/null
+++ b/ext/json/tests/json_encode_sse2_fuzz.phpt
@@ -0,0 +1,57 @@
+--TEST--
+json_encode()/json_decode() round-trip across many lengths (SSE2 fast path)
+--FILE--
+<?php
+/* The SSE2 fast path in php_json_escape_string() scans fixed 16-byte
+ * chunks, so a string containing a mix of clean ASCII, escapable ASCII,
+ * and multi-byte UTF-8 sequences will place its "interesting" bytes at
+ * every possible offset relative to a chunk boundary as its length
+ * varies. Round-tripping through json_decode() is a strong, deterministic
+ * oracle here since the decoder is untouched by this patch. */
+
+$alphabet = [
+    'a', 'b', 'c', ' ', '"', '\\', '/', '<', '>', '&', '\'',
+    "\t", "\n", "\r", "\x01",
+    "\xc3\xa9",         // 2-byte UTF-8 (e)
+    "\xe2\x82\xac",     // 3-byte UTF-8 (EUR SIGN)
+    "\xf0\x9f\x98\x80", // 4-byte UTF-8 (surrogate pair on encode)
+];
+$na = count($alphabet);
+
+$optionSets = [
+    0,
+    JSON_UNESCAPED_SLASHES,
+    JSON_UNESCAPED_UNICODE,
+    JSON_HEX_TAG | JSON_HEX_AMP | JSON_HEX_APOS | JSON_HEX_QUOT,
+];
+
+$failures = [];
+for ($count = 0; $count <= 80; $count++) {
+    $s = '';
+    for ($i = 0; $i < $count; $i++) {
+        // Deterministic index (no RNG) so the corpus is reproducible.
+        $idx = ($count * 31 + $i * 17) % $na;
+        $s .= $alphabet[$idx];
+    }
+
+    foreach ($optionSets as $opts) {
+        $encoded = json_encode($s, $opts);
+        if ($encoded === false) {
+            $failures[] = "count=$count opts=$opts: encode failed";
+            continue;
+        }
+        $decoded = json_decode($encoded);
+        if ($decoded !== $s) {
+            $failures[] = "count=$count opts=$opts: round-trip mismatch";
+        }
+    }
+}
+
+if ($failures) {
+    echo implode("\n", $failures), "\n";
+} else {
+    echo "OK\n";
+}
+?>
+--EXPECT--
+OK
diff --git a/ext/json/tests/json_encode_sse2_options.phpt b/ext/json/tests/json_encode_sse2_options.phpt
new file mode 100644
index 000000000000..1fd962287075
--- /dev/null
+++ b/ext/json/tests/json_encode_sse2_options.phpt
@@ -0,0 +1,75 @@
+--TEST--
+json_encode() option flags on strings long enough to hit the SSE2 fast path
+--FILE--
+<?php
+/* Same option-flag matrix as ext/json/tests/006.phpt and
+ * json_encode_unescaped_slashes.phpt, but padded past 16 bytes on both
+ * sides so the SSE2 fast path in php_json_escape_string() actually
+ * engages before and after the special character (short strings never
+ * exercise it). */
+
+function check($label, $actual, $expected) {
+    if ($actual === $expected) {
+        echo "$label: OK\n";
+    } else {
+        echo "$label: FAIL\n";
+        var_dump($expected, $actual);
+    }
+}
+
+$pad = str_repeat('x', 20);
+
+check('tag default', json_encode($pad . '<foo>' . $pad),
+    '"' . $pad . '<foo>' . $pad . '"');
+check('tag HEX_TAG', json_encode($pad . '<foo>' . $pad, JSON_HEX_TAG),
+    '"' . $pad . "\\u003Cfoo\\u003E" . $pad . '"');
+
+check('apos default', json_encode($pad . "'bar'" . $pad),
+    '"' . $pad . "'bar'" . $pad . '"');
+check('apos HEX_APOS', json_encode($pad . "'bar'" . $pad, JSON_HEX_APOS),
+    '"' . $pad . "\\u0027bar\\u0027" . $pad . '"');
+
+check('quot default', json_encode($pad . '"baz"' . $pad),
+    '"' . $pad . '\"baz\"' . $pad . '"');
+check('quot HEX_QUOT', json_encode($pad . '"baz"' . $pad, JSON_HEX_QUOT),
+    '"' . $pad . "\\u0022baz\\u0022" . $pad . '"');
+
+check('amp default', json_encode($pad . '&blong&' . $pad),
+    '"' . $pad . '&blong&' . $pad . '"');
+check('amp HEX_AMP', json_encode($pad . '&blong&' . $pad, JSON_HEX_AMP),
+    '"' . $pad . "\\u0026blong\\u0026" . $pad . '"');
+
+check('slash default', json_encode($pad . 'a/b' . $pad),
+    '"' . $pad . 'a\/b' . $pad . '"');
+check('slash UNESCAPED_SLASHES', json_encode($pad . 'a/b' . $pad, JSON_UNESCAPED_SLASHES),
+    '"' . $pad . 'a/b' . $pad . '"');
+
+check('unicode default', json_encode($pad . "\xc3\xa9" . $pad),
+    '"' . $pad . "\\u00e9" . $pad . '"');
+check('unicode UNESCAPED_UNICODE', json_encode($pad . "\xc3\xa9" . $pad, JSON_UNESCAPED_UNICODE),
+    '"' . $pad . 'é' . $pad . '"');
+
+check('lineterm default', json_encode($pad . "\xe2\x80\xa8" . $pad),
+    '"' . $pad . "\\u2028" . $pad . '"');
+check('lineterm UNESCAPED_UNICODE', json_encode($pad . "\xe2\x80\xa8" . $pad, JSON_UNESCAPED_UNICODE),
+    '"' . $pad . "\\u2028" . $pad . '"');
+check('lineterm UNESCAPED_UNICODE|UNESCAPED_LINE_TERMINATORS',
+    json_encode($pad . "\xe2\x80\xa8" . $pad, JSON_UNESCAPED_UNICODE | JSON_UNESCAPED_LINE_TERMINATORS),
+    '"' . $pad . "\xe2\x80\xa8" . $pad . '"');
+?>
+--EXPECT--
+tag default: OK
+tag HEX_TAG: OK
+apos default: OK
+apos HEX_APOS: OK
+quot default: OK
+quot HEX_QUOT: OK
+amp default: OK
+amp HEX_AMP: OK
+slash default: OK
+slash UNESCAPED_SLASHES: OK
+unicode default: OK
+unicode UNESCAPED_UNICODE: OK
+lineterm default: OK
+lineterm UNESCAPED_UNICODE: OK
+lineterm UNESCAPED_UNICODE|UNESCAPED_LINE_TERMINATORS: OK