Puppeteer vs TestDriver

Free and open source

From $20 per seat/month (120 testing minutes included, then $0.14/minute)

Free and open source. You pay in engineering time and CI minutes.

The seat price is the small half of the bill: $20 per seat covers 120 testing minutes, then metering kicks in at $0.14 a minute (about $8.40 an hour). Seats are billed for every GitHub user who gets a PR review or runs a test, and a vision model driving a desktop is slower per test than a headless browser — model the minutes before committing a large suite.

Everything, self-hosted

Free trial rather than a standing free tier; paid seats include 120 testing minutes a month.

Widely used

Google · since 2017

Niche

TestDriver

• Small, fast, and very well documented for Chrome-specific work.
• First-class access to the Chrome DevTools protocol for performance tracing and coverage.
• Excellent at non-test jobs: PDF rendering, screenshots, scraping.

• Tests surfaces nothing else reaches — browser extensions, packaged desktop apps, canvas and third-party OAuth screens.
• No selectors to maintain, so a refactor that renames every class breaks nothing.
• Runs per pull request in a disposable sandbox rather than against a shared staging environment.

• No test runner, assertions, retries or reporting — you assemble all of it.
• Chrome and Firefox only; no WebKit.
• For E2E testing specifically, Playwright is a strict superset. Choose Puppeteer only when you want a minimal automation library.

• A vision model is non-deterministic. Reruns of an unchanged test do not always take the same path, which makes a genuine failure harder to distinguish from a bad frame.
• Billed by testing hours, and driving a real desktop is slow — a suite that costs pennies on Playwright can cost real money here.
• Young product with a small community; expect to be an early reporter of your own bugs.

Automation jobs that are not tests, or Chrome-only tooling where you want minimal dependencies.

Teams testing extensions, desktop apps or canvas-heavy UIs, where selector-based runners simply cannot see the thing under test.