Performance

Benchmarks

Editor and reader measurements from the recorded August 27, 2026 run. Compare the workload and feature differences alongside the numbers.

Hard test

500 blocks on a 4× slower CPU

Cold p50. Lower is better.

System	Ready ms	Type p50 / p95 ms	Change	Long tasks ms
WidgetsPage builder	172	77 / 79	2.1 KB	59
PuckPage builder	1,159	785 / 787	98 KB	3,211
BlockNoteText blocks	353	195 / 197	706 B	288
Craft.jsPage builder	1,288	250 / 253	172 KB	5,407
GrapesJSPage builder	1,181	37 / 39	214 KB	1,743
LexicalText reference	145	60 / 61	161 KB	59
Plain ReactHand-built reference	138	136 / 139	89 KB	0

Widgets opens 6.7× faster than Puck. GrapesJS types faster. BlockNote sends the smallest change. Lexical and plain React are narrower text references, not mixed page builders.

Normal test

100 blocks on the native CPU

Cold p50, with typing p95.

System	Ready ms	Heap MiB	DOM	Type p50 / p95 ms	Change
Plain ReactEditor	41	1.76	565	11.7 / 12.0	17.6 KB
Plain ReactViewer	46	1.76	565	—	—
LexicalEditor	46	2.36	215	10.0 / 30.3	32.2 KB
WidgetsEditor	47	2.23	665	8.9 / 32.0	2.1 KB
WidgetsViewer	47	1.95	665	—	—
PuckEditor	268	10.90	907	69.7 / 71.4	19.5 KB
PuckViewer	63	2.84	566	—	—
BlockNoteEditor	78	4.10	421	42.5 / 43.8	706 B
Craft.jsEditor	75	4.54	565	37.2 / 38.1	34.4 KB
GrapesJSEditor	125	13.22	132*	6.6 / 6.9	43.0 KB

The GrapesJS DOM count excludes its iframe canvas. The viewer rows have no input or change callback by design.

Downloads

Production payload, gzip KiB

System	Initial	Lazy JS
Plain ReactEditor	61.2	0
Plain ReactViewer	60.9	0
LexicalEditor	133.5	0
WidgetsEditor	68.1	100.9
WidgetsViewer	63.5	0
PuckEditor	242.4	119.7
PuckViewer	188.3	120.4
BlockNoteEditor	338.0	108.0
Craft.jsEditor	90.4	0
GrapesJSEditor	312.4	0

Published pages

Viewer-only cost

Renderer	Initial	100 ready	500 at 4×	Heap
Plain React	60.9 KB	46 ms	129 ms	1.76 MiB
Widgets	63.5 KB	47 ms	142 ms	1.95 MiB
Puck Render	188.3 KB	63 ms	189 ms	2.84 MiB

Heap is the 100-block native result.

GrapesJS wins direct typing.37 ms versus 77 ms for Widgets in the hard test.

BlockNote sends less.706 B for its changed text block versus a 2.1 KB widget patch.

The slow tail is visible.At 100 native blocks, Widgets measured 8.9 ms p50 and 32 ms p95.

Payloads are not identical.Widgets and BlockNote send changed content. The other editors send their normal full state.

Measured

What the 600 runs cover

Initial and lazy downloadsCold and warm ready timeIdle and active heapDOM element countEditor activation20-character typing2,000-character pasteChange payload sizeJSON serializationMarkdown exportReplace and selectInsert, move, copy, deleteUndo and redoScroll framesBrowser long tasks

Evidence

Read or download the run

Readable report 5 KB Methodology 4 KB p50 and p95 JSON 774 KB All 600 samples 1.2 MB

Exact setup and comparison limits

Run 2026-08-27T17-52-52.090 used ten isolated Vite production builds on a 10-core Apple M4 with 16 GiB RAM. The matrix covers 25, 100, and 500 blocks; cold and warm loads; 1× and 4× CPU; and five repetitions. The full matrix produced 600 successful samples with no recorded benchmark errors.

Puck and GrapesJS include substantial editor UI. Craft.js and Widgets are lower-level page systems. BlockNote and Lexical are text-first. Plain React is a hand-built reference. These are product architectures, not feature-identical editor cores.