<Flashwright />

The fastest e2e framework.

A drop-in Playwright: same API, same test runner, same npx playwright test. Your suite runs unmodified, 2–4× faster than on Playwright and faster than on every framework we benchmarked.

$ npm i -D @playwright/test@npm:flashwright-test
$ npx playwright test
  24 passed (1.1s)

// Side by side

Same 24 tests, under a third of the time

Playwright's own TodoMVC example suite, run by both on the same machine. Real output, played back in real time.

~/todomvc

// Compared

Faster than every framework we measured

The same 24 TodoMVC tests ported to 9 tools, on the same machine and the same Chromium. Each tool uses its own recommended waits and isolation, and every counted run passes all 24 tests. Median of 15 runs of the whole test command.

1 worker

Flashwright
fastest
2.36 s
Bun test + Bun.WebView
1.8×
4.30 s
Playwright
1.9×
4.48 s
Selenium + Mocha
2.3×
5.40 s
Puppeteer + Jest
2.6×
6.17 s
Nightwatch
2.7×
6.29 s
WebdriverIO
4.4×
10.32 s
Cypress
7.2×
16.97 s
TestCafe
9.5×
22.46 s

4 workers

Flashwright
fastest
1.47 s
Bun test + Bun.WebView
1.8×
2.67 s
Selenium + Mocha
1.8×
2.70 s
Puppeteer + Jest
2.0×
2.88 s
Playwright
2.0×
3.02 s
Nightwatch
2.2×
3.25 s
WebdriverIO
3.2×
4.74 s
TestCafe
6.2×
9.16 s
Cypress
11.5×
16.99 s

Cypress runs one spec file at a time: parallel runs need Cypress Cloud.

Tests grouped by feature in 7 files. With one test per file, tools that start something per file fall further behind. Every port, version and setting is in bench/competitors; corrections are welcome.

// Benchmarks

At least 2× on every scenario

The whole test command, launch to exit. Median of 15 alternating runs, same machine, Chromium headless shell, warm session off.

Playwright 1.64.1Flashwright
TodoMVC example suite 24 tests · 1 worker
3.74×
3617 ms
968 ms
TodoMVC example suite 24 tests · 4 workers
2.45×
1767 ms
721 ms
SVGOMG example suite 6 tests · 1 worker
3.23×
1687 ms
523 ms
SVGOMG example suite 6 tests · 4 workers
2.68×
1242 ms
463 ms
Shop app with a sign-in setup project 18 tests · 1 worker
2.18×
6429 ms
2950 ms
Shop app with a sign-in setup project 18 tests · 4 workers
2.15×
2558 ms
1189 ms
Admin app, 8,500 elements 19 tests · 1 worker
2.23×
6745 ms
3018 ms
Admin app, 8,500 elements 19 tests · 4 workers
2.04×
2639 ms
1293 ms
Trivial tests 15 tests
2.78×
975 ms
351 ms
Click + assertion 15 tests
4.10×
1582 ms
386 ms
Assertion on a delayed update 15 tests
2.44×
1156 ms
474 ms
Navigation 15 tests
2.89×
1213 ms
419 ms
Fixture setup only 15 tests
3.21×
1138 ms
355 ms
Form submit to a backend 15 tests
2.67×
1910 ms
715 ms
toHaveScreenshot() 15 tests
2.51×
2297 ms
916 ms
page.screenshot() 15 tests
2.48×
1867 ms
752 ms
Suite with real failures 15 tests
3.78×
4266 ms
1128 ms
Expected failures 15 tests
3.16×
4735 ms
1500 ms
One test, as an agent re-runs it 1 test
3.09×
577 ms
187 ms

Reproduce with node bench/matrix.mjs 15. The full table, and how much of each run is the app's own waiting, is in the README.

// How it works

Where the time goes

Flashwright changes only where Playwright spends time. Playwright's own test suites are the regression bar.

01

Warm browsers, fresh contexts

Each test gets a new tab in a context the worker keeps, with everything a test can change reset in between.

02

Waits that react to the page

Actions and assertions re-check when the page changes, not on a backoff timer.

03

Input without frame waits

Clicks don't wait for animations to settle. The hit-target check still runs.

04

Workers started early

Workers and browsers launch while the config is still being read. The longest tests start first.

05

A faster start

A V8 code cache, lazy loading and fewer browser round trips per action.

06

Faster screenshots

Same pixels, faster encoding, and baselines byte-identical to Playwright's.

// For agents

Re-runs in 72 ms

Coding agents re-run the same test after every edit. Outside CI, a warm session keeps the runner, workers and browsers ready for the next run.

Re-runPlaywrightFlashwrightWarm session
One TodoMVC test652 ms234 ms72 ms
TodoMVC, 4 workers1751 ms710 ms490 ms
mcp
$ playwright mcp --adaptive-settle
# answers once the page is quiet,
# not after a fixed 500 ms

// Options

Fast by default, Playwright on request

Every default that changes behavior is a typed option. Turn Playwright's behavior back on per project, file or describe block.

playwright.config.ts
export default defineConfig({
  testOrder: 'file',
  use: {
    isolation: 'strict',
    waitForAnimations: true,
    failFastWhenIdle: false,
  },
});
OptionDefault
isolation'recycled' · a new tab in a reset context per test
failFastWhenIdletrue · fail once nothing can change the page
waitForAnimationsfalse · don't wait for moving elements
testOrder'longest-first' · slowest tests start first
locatorPolicy'any' · 'test-id' allows only getByTestId()

More, like workersPerProcess, preloadPages, instantAnimations and blockResources, in the README.

// Questions

Before you switch

Do I have to change my tests?

No. Imports, configs and npx playwright commands stay as they are. Browsers, reporters, the trace viewer and the VS Code extension are Playwright's.

What behaves differently?

Contexts are recycled, pointer actions don't wait for animations, idle assertions fail early, and the longest tests run first. Each has an option to restore Playwright's behavior. The README lists every difference.

Which browsers get faster?

Mostly Chromium. Firefox and WebKit keep Playwright's behavior, apart from reusing one tab per worker. Flashwright runs Playwright's own browser builds.

How do I know nothing broke?

Playwright's own test suites are the bar: tests/page passes 2610/2610 on Chromium, and the runner, library and MCP suites match pristine Playwright 1.64.1.

How do I go back?

npm i -D @playwright/test@1.64.1. Nothing in your code refers to Flashwright.