Playwright now takes 84.0% of weekly npm downloads among the 11 browser testing tools we track. Puppeteer has 8.0%, Cypress 4.1%, WebdriverIO 1.7% and Selenium 1.1%. That's the week ending 8 November 2026, from api.npmjs.org, fetched 14 November 2026.
A year earlier Playwright had 59.0%. In late August it had 75.8%. Apart from 2 small dips (December 2025 and February 2026), it has gained share every month for the past year, and the pace picked up this autumn instead of slowing down.
The chart shows share, not raw downloads. Raw npm counts have inflated across the whole registry, so a line of raw totals would mostly show registry growth. The methodology section below explains the correction.
Five tools have led in 11 years#
Counting only weeks where the 11 tools together passed 150,000 downloads, the lead changed hands like this (5-week centred average):
| From | Leader | Share when it took the lead |
|---|---|---|
| March 2015 | Selenium (selenium-webdriver) | 51.4% |
| May 2016 | PhantomJS (phantomjs-prebuilt) | 32.2% |
| July 2017 | Selenium again | 34.1% |
| November 2020 | Puppeteer | 20.9% |
| May 2021 | Cypress (traded the lead with Puppeteer through 2021 and again in late 2023) | 22.3% |
| May 2024 | Playwright | 27.1% |
Playwright took the top spot in the week ending 26 May 2024 with 27.12% against Cypress's 27.08%. Six months later it had 40.0%. It hasn't been close since.
Snapshot each November#
| Week ending | Playwright | Cypress | Puppeteer | Selenium |
|---|---|---|---|---|
| 5 Nov 2023 | 14.2% | 28.3% | 29.4% | 9.9% |
| 10 Nov 2024 | 40.0% | 22.7% | 16.2% | 7.3% |
| 9 Nov 2025 | 59.0% | 14.9% | 14.0% | 3.9% |
| 8 Nov 2026 | 84.0% | 4.1% | 8.0% | 1.1% |
Cypress set a download record and still lost share#
Cypress's weekly downloads hit their all-time high in late August 2026, about 7.56 million a week on the 5-week average. Its share of the category peaked much earlier, at 31.6% in July 2023. Both things are true because the registry kept growing around it.
Compare January 2025 with the 4 weeks to 8 November 2026. Two control packages whose real usage barely changed grew anyway: lodash by 3.47x and typescript by 5.91x. Over the same span playwright grew 12.49x, cypress 1.32x and selenium-webdriver 1.22x. Measured against registry inflation, Cypress and Selenium both lost ground in real terms while their raw numbers went up.
Selenium's npm number undercounts Selenium#
Treat the 1.1% with care. Most Selenium code is Java, Python or C#, and it never touches npm. This chart only measures the JavaScript slice, where Selenium was never as strong as it is in enterprise Java estates.
Stack Overflow shows the other side. The selenium-webdriver tag (which now absorbs the old selenium tag as a synonym) has 57,350 questions. playwright has 3,502, cypress 9,695 and puppeteer 8,001 (Stack Exchange API, fetched 14 November 2026). That's 16x more Selenium questions than Playwright ones. Years of answers and existing suites don't move just because new projects pick something else.
Not every Playwright download is a test#
The playwright package pulled 166,770,617 downloads in the week ending 8 November 2026. Some of that isn't testing:
@playwright/test, the test runner, depends onplaywrightand pulled 117,911,590 downloads the same week. Those installs are almost all test suites.@playwright/mcp, the server that lets AI agents drive a browser, also depends onplaywrightand pulled 10,233,094.- The remainder includes scrapers, automation scripts and agent tools that use the library directly.
The autumn jump is part of this. From the week ending 30 August to the week ending 8 November 2026, playwright went from 87.5 million to 166.8 million weekly downloads (1.91x). typescript grew 1.53x over the same 10 weeks and lodash 1.18x, so most of the jump is real, but npm can't tell you how much came from agents rather than test suites.
Puppeteer has the same caveat in a stronger form. A large share of its downloads are scraping and PDF generation, not testing. It stays in the basket because people do still use it for browser tests, and dropping it would hide the 2020 to 2023 stretch when it led.
Full standings#
Week ending 8 November 2026, api.npmjs.org, fetched 14 November 2026:
| Package | Downloads that week | Share (5-week average) |
|---|---|---|
playwright | 166,770,617 | 84.04% |
puppeteer | 15,279,663 | 8.05% |
cypress | 7,695,243 | 4.11% |
webdriverio | 3,318,462 | 1.69% |
selenium-webdriver | 2,104,005 | 1.12% |
protractor | 615,449 | 0.33% |
codeceptjs | 333,176 | 0.22% |
phantomjs-prebuilt | 398,383 | 0.22% |
testcafe | 214,352 | 0.12% |
nightwatch | 123,593 | 0.09% |
nightmare | 15,632 | 0.01% |
Protractor's own npm deprecation notice says it reached end of life in summer 2023. It still pulls over 600,000 downloads a week, so a lot of CI pipelines are still installing it.
Methodology#
Daily download counts for the 11 packages above came from the public api.npmjs.org/downloads/range endpoint, 10 January 2015 through 13 November 2026, in chunks of about 17 months to stay under the API's 18-month limit. Fetched 14 November 2026. Days were summed into weeks ending Sunday, and only complete 7-day weeks were kept (617 weeks, 18 January 2015 to 8 November 2026).
Three judgement calls:
playwright, not@playwright/test. The test runner depends onplaywright, so counting both would count most Playwright installs twice. No other tool in the basket has the same problem at this scale.- Share, not raw counts. Every package's raw weekly count was smoothed with a 5-week centred rolling mean, then divided by the basket total for that week. Registry-wide inflation hits every package in the basket, so it cancels out of the share.
lodashandtypescriptwere pulled as controls only and are not in the basket. - No spike removal. Unlike our Claude Code vs Codex report, no single-package scraper spike stood out here. The week ending 13 September 2026 dipped for every package at once (
playwright69.4 million against 98.6 million the week before), which looks like a registry counting gap rather than a real drop. The 5-week average smooths it and the share barely moves.
The CSV below has the raw weekly counts and the computed share for every package and every week. The chart samples every fourth week.
ObserveOne, "E2E Testing Tools by npm Downloads: 11 Years of Share," fetched 14 November 2026. Full weekly dataset (CSV): npm-share-e2e-testing-tools.csv. Source: api.npmjs.org.
What we make of it#
We build ObserveOne on Playwright, so we aren't neutral. The download share says the framework question is mostly settled for new JavaScript projects. If you want the head-to-head detail, see Playwright vs Cypress and Playwright vs Selenium.
The harder question is the one the downloads can't answer: who keeps those tests honest after the first month. A suite that goes flaky and gets skipped fails the same way on any framework. That's the problem we work on, starting with why Playwright tests turn flaky and how to fix them.