End-to-end browser tests for a small web app: Playwright Test, Cypress and WebdriverIO

Playwright Test is the most complete free starting point for a small team’s end-to-end tests. WebdriverIO is the choice when you must test real Safari; Cypress suits teams already invested in it.

PRACTICAL PICKSPreferredPickOur Preferred PickA new suite for your own web app$0

Playwright Test

One free package with the runner, a fresh browser context for every test, Chromium, Firefox and WebKit, and a local trace viewer for failed tests.

Its WebKit and Firefox are patched builds, not branded Safari or Firefox. Treat saved traces as sensitive.

Why Playwright Test is our Pick
Newcomer assessment

No Rising Star selected

On October 24, 2026 we screened launches since 2024, including TesterArmy e2e, OpenCheck and Bombadil, alongside the September 2026 Biloba client.

Each failed on platform coverage, pre-1.0 status, AI-dependent execution or a different testing model. None earned an endorsement for this reader.

Who is this article for? Small teams writing end-to-end browser tests in JavaScript or TypeScript for their own web app, run locally and in CI, who want to choose a framework before the suite grows.

About 24 min read

What Preferred Pick, Rising Star and Ready to Rock mean

The quick list

All three test runners are free. Compare browser coverage, how each isolates tests and what you keep when a test fails.

All three test runners are free for local and CI use under our October 24, 2026 cost rule pass. Cypress Cloud is optional: Starter is free for 500 recorded results a month; Team is $799 a year or $75 a month in US dollars, before any tax. CI machines and maintenance time cost money with any of the three.

The comparison at a glance

Compare browser coverage, how each framework isolates tests, what you keep when a test fails and the 2026 changes your suite will meet.

Order follows our verdict for a new suite on a small web app, not measured speed, reliability or security. All tools remain selectable.

JavaScript and TypeScript end-to-end browser test frameworks for a small web app
FeaturePlaywright TestCypress AppWebdriverIO
Best fitPreferred Pick: a new JS or TS suite for your own web app, run locally and in CIA suite that already runs on Cypress, with a team that relies on its Command LogAlso consider: real Safari, or existing WebDriver, Grid or Appium infrastructure
Test execution modelIntegrated runner, assertions, reports and debugging tools; JS and TSCommands run inside the browser; Node tasks for server workNode.js workers with Mocha, Jasmine or Cucumber adapters
Browser targetsChromium, patched Firefox and WebKit; installed Chrome and EdgeLatest three Chrome, Firefox (140+) and Edge majors; experimental WebKitBranded Chrome, Firefox, Edge and Safari through standard drivers
Browser-state isolationNew browser context per test by defaultOn by default: page, cookies, local and session storage; IndexedDB persistsBrowser session per spec file and browser configuration
Saved failure recordLocal trace ZIP: DOM snapshots, screencast, console, network headers and bodiesTest Replay needs Cloud recording; not on experimental WebKitDevTools trace.zip for offline replay
Change to plan forRoughly monthly releases; each pins new browser builds to reinstall16.0 (September 1, 2026) removed Cypress.env(), cy.exec() and cy.end(); Node.js 20 dropped10.0 (October 5, 2026): W3C-only sessions, strict $, removed executeAsync and other old commands
Paid service costNone needed for local or CI runs; hosted services not evaluatedStarter free; Team $799/year or $75/month (USD); taxes not statedNone needed; third-party browser grids priced separately, not evaluated

Public-source research checked October 19, 2026 for capabilities and October 24, 2026 for cost rule v1 and Rising Star screening, for a small team writing JavaScript or TypeScript end-to-end tests for its own web app without a required paid service. We installed nothing and ran no tests. Native mobile, component, load, visual-regression and AI-generated testing are out of scope. Unknown means unresolved, not absent. Download and star counts measure community size, not quality. Next substantive review: December 18, 2026, or sooner after a major release.

All features and sources 21 criteria

Open the groups that matter to you. They compare the same selected tools. “Planned” means future work, not an available feature. Sources include the published conditions.

Runner, browsers and test state

Show 8 criteria
Runner, browsers and test state
FeaturePlaywright TestCypress AppWebdriverIO
Best fitPreferred Pick: a new JS or TS suite for your own web app, run locally and in CIA suite that already runs on Cypress, with a team that relies on its Command LogAlso consider: real Safari, or existing WebDriver, Grid or Appium infrastructure
Local runner licenseApache 2.0; no chargeMIT; free app, optional paid CloudMIT; no charge; OpenJS Foundation project
Test execution modelIntegrated runner, assertions, reports and debugging tools; JS and TSCommands run inside the browser; Node tasks for server workNode.js workers with Mocha, Jasmine or Cucumber adapters
Browser targetsChromium, patched Firefox and WebKit; installed Chrome and EdgeLatest three Chrome, Firefox (140+) and Edge majors; experimental WebKitBranded Chrome, Firefox, Edge and Safari through standard drivers
Safari qualificationWebKit build only; does not work with branded SafariExperimental WebKit; no cy.origin(), Test Replay or network-error simulation thereReal Safari through SafariDriver, which means a Mac
Browser-state isolationNew browser context per test by defaultOn by default: page, cookies, local and session storage; IndexedDB persistsBrowser session per spec file and browser configuration
Separate users or tabsSeveral contexts in one testOne browser at a time; @cypress/puppeteer plugin for tabsMultiremote: several named browser sessions in one test
Network substitutionRoute fulfillment replaces or modifies responsescy.intercept() stubs; native interception in Chrome-family browsers since 16.0mock requires WebDriver BiDi: usually local Chromium or Firefox; Safari not stated
Sources for runner, browsers and test state

Checked October 19, 2026: Playwright introduction, browsers, contexts, best practices and license; Cypress license, trade-offs, browsers, isolation, configuration and changelog; WebdriverIO overview, runner, protocols, mocks, v10 migration guide and license. Browser-state isolation does not reset a shared backend. No browser compatibility was tested by us.

Debugging, results and upkeep

Show 8 criteria
Debugging, results and upkeep
FeaturePlaywright TestCypress AppWebdriverIO
Interactive inspectionUI mode with watch and time travel; InspectorCommand Log snapshots and browser developer toolsOptional DevTools service: browser UI, per-command screenshots, single-test reruns
Saved failure recordLocal trace ZIP: DOM snapshots, screencast, console, network headers and bodiesTest Replay needs Cloud recording; not on experimental WebKitDevTools trace.zip for offline replay
Parallel executionParallel workers; sharding across machinesParallel CI runs through Cloud, free tier included; CI machines separateConcurrent workers per spec file and browser
Local result filesHTML report and trace ZIPBundled JUnit XML; give each spec its own file nameOptional JUnit XML reporter, one file per spec
Current release, checked October 19, 20261.64.1, October 14, 2026; docs list Node.js 22, 24 or 2616.2.0, October 2026; Node.js 22, 24, or 26 and later10.1.0, October 18, 2026; Node.js 22.19 or later
Change to plan forRoughly monthly releases; each pins new browser builds to reinstall16.0 (September 1, 2026) removed Cypress.env(), cy.exec() and cy.end(); Node.js 20 dropped10.0 (October 5, 2026): W3C-only sessions, strict $, removed executeAsync and other old commands
Community size, mid-October 2026About 37.7M npm downloads a week; 98k GitHub stars; Microsoft projectAbout 2.8M a week; 51k stars; Cypress.io, Inc.About 1.3M a week; 9.8k stars; OpenJS Foundation
Security recordOne npm advisory: macOS browser reinstall scripts, fixed in 1.55.1 (CVE-2025-59288)No GitHub advisory; August 2026 Cloud data incident, final forensic result unpublishedBrowserStack service command injection, fixed in 9.24.0 (CVE-2026-25244)
Sources for debugging, results and upkeep

Playwright introduction, trace viewer, release notes, npm record and CVE-2025-59288; Cypress features, CI strategies, reporters, changelog, npm record and incident disclosure; WebdriverIO DevTools, DevTools limitations, runner, JUnit, v10 release, npm record and CVE-2026-25244. Downloads are npm counts for October 12 to 18, 2026 and include CI and dependent installs; they show community size, not quality. WebdriverIO DevTools minimum release and browser parity remain unresolved.

Optional paid services, retained data and AI features

Show 5 criteria
Optional paid services, retained data and AI features
FeaturePlaywright TestCypress AppWebdriverIO
Paid service costNone needed for local or CI runs; hosted services not evaluatedStarter free; Team $799/year or $75/month (USD); taxes not statedNone needed; third-party browser grids priced separately, not evaluated
Cloud quantity and usersNo recorded-result allowanceStarter: 500 results/month, 10 users; Team: 120,000/year, 50 usersNo recorded-result allowance
What a paid meter countsNothing; no result subscriptionRecorded it() runs; retries count once; local and skipped tests excluded; extra results $6 per 1,000 on TeamNothing; no result subscription
History and sensitive artifactsYour local or CI storage; traces can hold credentials in requestsCloud history 30 days free, 90 days Team; follow the incident advice on secretsYour storage; masking does not cover every exposure
AI agent features (not scored)Optional test agents for Claude Code, VS Code and others; MCPMetered cy.prompt; cypress tap; Cloud MCP for Claude and other clientsOptional AI service (Anthropic, OpenAI or local models); MCP; agent skill
Sources for optional paid services, retained data and ai features

Cypress Cloud pricing and retention, checked October 19, 2026; the $75 monthly Team price comes from the page’s published price data. Cypress privacy and August 2026 disclosure. Cloud quotas do not limit local runs. The separate Cloud terms were not readable by our automated check, so refunds and early cancellation remain unresolved. Playwright trace contents and WebdriverIO logging limitations inform artifact handling. AI features from Playwright test agents, Cypress Cloud MCP and the WebdriverIO v10 release carried no weight. No paid service was used.

What matters when choosing

Playwright Test is our Practical Pick for a small team starting an end-to-end browser test suite for its own web app in JavaScript or TypeScript, run locally and in CI without a paid service. It is free under the Apache 2.0 license and bundles the parts a small team would otherwise assemble: a test runner, retrying assertions, a fresh browser context for every test, parallel workers, an HTML report, an interactive UI mode and a trace viewer that replays a failed test from a saved file. One configuration can run the same tests in Chromium, Firefox and WebKit, and one test can open separate contexts for two users, such as the sender and the recipient of an invitation. Disclosure: Practical Picks is researched and written with Anthropic’s Claude; Playwright, WebdriverIO and Cypress all advertise optional Claude integrations, which carried no weight in our judgment. Introduction, isolation, trace viewer, license, test agents

The most important compromise is browser coverage. Playwright’s Firefox and WebKit are patched builds, not the Firefox and Safari your visitors install, and its documentation says it does not work with branded Safari; WebKit on macOS is the closest it offers. Each Playwright version also pins its own browser builds, so an upgrade means reinstalling browsers. Traces can hold request and response headers and bodies, so treat a failure record from a signed-in test as sensitive. Browsers, trace contents

Choose WebdriverIO instead if you must automate real Safari or keep existing WebDriver, Selenium Grid or Appium infrastructure. It drives branded Chrome, Firefox, Edge and Safari through their standard drivers and is run by its community under the OpenJS Foundation. You assemble more of the setup, each spec file rather than each test gets a fresh browser session, and version 10, released October 5, 2026, removed older commands and requires Node.js 22.19 or later. If your suite already runs on Cypress, staying can be reasonable, because its in-browser Command Log is a documented strength. Plan for Cypress 16, which on September 1, 2026 removed Cypress.env() and cy.exec(). Cypress controls one browser at a time, treats WebKit as experimental and in August 2026 disclosed a Cypress Cloud data incident whose final forensic result has not been published. WebdriverIO v10, Cypress changelog, Cypress trade-offs, incident disclosure

All three test runners are free for this job. On October 24, 2026 we re-scored cost under rule v1 for a three-year horizon without a required paid service; every fit runner priced at $0, so Playwright Test still totals 4.38 at slopes 1, 1.5 and 2, ahead of WebdriverIO (3.53) and Cypress App (3.35). Paid services stay optional: Cypress Cloud’s Team plan costs $799 a year if you record runs, and your CI machines cost money with any framework. These are researched judgments from documentation, release notes, policies and public reports checked October 19, 2026, with pricing re-read October 24, 2026. We did not install or run any framework, and we name no speed, reliability or security winner. Whichever you choose, start with two journeys: an ordinary form submission and your most awkward interaction. Decide what must be real and what may be substituted. A mocked response can exercise the screen without proving that your live backend produces it, and a fresh browser context does not reset a shared database. Cypress pricing

Playwright Test

Preferred Pick · a new suite for your own web app

$0Apache 2.0 · runner, reports and traces

Choose Playwright Test for a new JavaScript or TypeScript suite when you want the runner, per-test isolation, three browser engines and a local failure record in one free package. Its WebKit and Firefox are not the branded browsers.

Reasons to choose it

  • Runner, assertions, HTML report, UI mode and trace viewer included
  • New browser context per test; several users in one test
  • Chromium, Firefox and WebKit from one configuration

What to watch

  • Does not work with branded Safari or Firefox; patched builds instead
  • Each version pins its own browser builds
  • Traces can contain request headers and bodies

Playwright Test bundles “test runner, assertions, isolation, parallelization and rich tooling”, according to its introduction, and runs JavaScript or TypeScript tests headed or headless, locally or in CI. Its locators wait until an element can be acted on, and its assertions retry until the expected state appears, which removes many hand-written waits but does not prove that an order was saved correctly. The HTML report filters by browser and by passed, failed, skipped or flaky tests, and UI mode adds watch mode and step-by-step time travel. We verified that version 1.64.1, released October 14, 2026, is Apache 2.0 licensed. The documentation lists Node.js 22, 24 or 26, Windows 11 or later, macOS 14 or later and recent Debian and Ubuntu releases. Introduction, best practices, package record

Its strongest reason not to choose it is that it cannot test the Safari or Firefox your visitors install, only patched builds of their engines. Choose WebdriverIO if branded Safari or existing driver infrastructure matters more than an integrated toolset, and stay with Cypress if your team already depends on its in-browser workflow.

Read the full Playwright Test analysis and sources

Browser contexts are not database resets

Playwright Test creates a new browser context for every test by default, and one test can create several contexts, for example an administrator and an ordinary user in the same scenario. That suits an invitation flow seen from both sides. Both users may still change the same server-side record, so plan how test data is created and removed; an isolated browser cannot repair a shared database left in an unexpected state. Version 1.64, released October 7, 2026, added a --shuffle option that runs tests in random order, which the release notes say helps find tests that depend on each other. Isolation, release notes

Know what “WebKit” and “Firefox” mean here

Playwright runs open-source Chromium by default and can also run installed Chrome and Edge. Its Firefox and WebKit builds rely on patches, and the documentation says Playwright does not work with the branded versions of Firefox or Safari. Its WebKit comes from WebKit’s main branch, often ahead of Safari, and media codecs “vary substantially” between Linux, macOS and Windows; the project suggests WebKit on macOS for the closest result to Safari. Device presets are emulation, not tests on physical phones. Version 1.64 added a default: false project setting, so a team can keep Firefox and WebKit configured but run them only when asked, for example in CI. Browsers, release notes

Keep a useful failure record without assuming it is harmless

A trace is a ZIP file with DOM snapshots around each action, a screencast, console logs, source locations and network requests, including headers and bodies. The documentation says the browser-based viewer “loads the trace entirely in your browser and does not transmit any data externally”, and recommends recording traces on the first retry of a failed test in CI. That statement covers the viewer, not your CI artifact storage or any service you upload to. A screenshot may show an account name and a request may carry an authorization value, so restrict who can download failure artifacts and how long they are kept. Trace viewer

Security record, privacy and AI features

OSV lists one advisory for the playwright package: before version 1.55.1, released in October 2025, macOS scripts that reinstall branded Chrome and Edge downloaded installers without checking the server’s certificate (CVE-2025-59288). We found no advisory for Playwright’s packages published since our September check. Microsoft’s general privacy statement, last updated September 2026, applies according to the product you use; it is not evidence that local tests are uploaded or used for training. We did not evaluate any paid Microsoft testing service. Playwright also documents optional AI “test agents” that plan, generate and repair tests through Claude Code, VS Code and other agent tools; they carried no weight in our judgment. CVE record, Microsoft privacy statement, test agents

Used Playwright Test? Share your experience to help other readers choose.

Cypress App

Existing Cypress suites · plan the 16 upgrade

$0MIT app · Cloud Team $799/yr

Stay with Cypress if your suite and team already rely on its in-browser Command Log. For a new suite, its one-browser model, experimental WebKit and August 2026 Cloud incident weigh against it.

Reasons to choose it

  • Command Log snapshots and browser developer tools while tests run
  • Isolation on by default for page, cookies and storage
  • Free Cloud tier with Test Replay and parallel CI runs

What to watch

  • One browser at a time; multiple tabs need a plugin
  • Cypress 16 removed Cypress.env() and cy.exec()
  • Cloud incident exposed some tokens and test code; final result unpublished

Cypress runs test commands inside the browser, beside your app. Its documentation describes a Command Log with snapshots you can step back through, browser developer tools, automatic waiting, spies, stubs and network controls. These make the sequence of actions easy to inspect; that is a documented capability, not a measured debugging advantage. The app is free and MIT-licensed. We verified that version 16.2.0 was released in October 2026 and requires Node.js 22, 24, or 26 or later. Why Cypress, pricing and license, package record

The strongest reason not to start a new suite on it is the architecture. “You cannot use Cypress to drive two browsers at the same time”; multiple tabs need the @cypress/puppeteer plugin, each test is bound to one site unless you use cy.origin(), iframe support is “somewhat limited”, and test code cannot import server-side modules, so database setup goes through cy.task() or cy.request(). If your key journey needs two users acting at once, Playwright’s multiple contexts are a more direct documented fit. Stay with Cypress if your suite already works well on it. Trade-offs

Read the full Cypress App analysis and sources

Browser support and isolation have limits

Cypress supports the latest three major versions of Chrome, Firefox and Edge and cannot launch Firefox older than 140. WebKit, Safari’s engine, is experimental: it needs a configuration flag and the playwright-webkit package, and cy.origin(), Cloud Test Replay and network-error simulation do not work there. The Electron browser is deprecated and is to be removed. Test isolation is on by default and clears the page, cookies and local and session storage before each test, but IndexedDB and other browser storage are not cleared. Launching browsers, test isolation, configuration

Plan for the Cypress 16 upgrade

Cypress 16.0.0, released September 1, 2026, removed Cypress.env() in favor of cy.env() and Cypress.expose(), removed cy.exec() in favor of cy.task(), removed cy.end() and dropped Node.js 20. Chrome, Chromium and Edge now send test traffic through native browser network interception, which changes a few cy.intercept() behaviors; Firefox and WebKit keep the older path. The release also turned on browser memory management by default and removed the 10 ms delay between typed keystrokes. Budget time for these changes if you maintain an older suite. Changelog

Local results and Cloud quotas are different

Cypress writes local JUnit XML without a Cloud account; give each spec its own file name, because a static name is overwritten by each spec in turn. Cypress Cloud is optional and records runs from CI for Test Replay, history, parallel runs and analytics. Its free Starter plan includes 10 users, 500 recorded test results a month and 30 days of history. Team costs $799 a year, which the page presents as “starting at $67 / month”; the page’s own price data also lists $75 a month on monthly billing. Team includes 50 users, 120,000 results a year and 90 days of history, with extra results at $6 per 1,000. Only results recorded with --record count, a retried test counts once and local runs do not count. Prices are shown in US dollars; taxes are not stated. Reporters, pricing

The pricing page says you can change or cancel a plan anytime. Cypress’s website terms say they do not govern Cloud, and the separate Cloud terms did not load as readable text for our automated check, so refunds and ending an annual plan early remain unresolved. Pricing, website terms

Review the August 2026 Cloud disclosure before recording

Cypress disclosed that an attacker exploited a previously unknown vulnerability in the Metabase analytics instance it used, on July 31, 2026. Its latest update, on August 17, says the exposed data included limited business and account data and repository and build metadata and, in a subset of cases, certain tokens and compiled test bodies and hooks. It says Test Replays, screenshots, videos, console logs, record keys, passwords and payment details were not affected. Cypress reports no evidence of ongoing unauthorized access after its preliminary investigation and says an independent forensic firm is reviewing its work; we found no published final result as of October 19. It advises rotating secrets placed in build parameters or recorded tests and long-lived tokens in repository URLs. This is not a finding that the local app was compromised, and we have not verified containment or how many customers were affected. Incident disclosure

Cypress.io, Inc.’s privacy policy, last updated October 17, 2024, separates information about its own site and service users from customer end-user data processed under service agreements. In 2023, Cypress acknowledged blocking certain third-party results dashboards, so check that any replacement results service is supported before relying on it. Cypress also sells metered AI test generation (cy.prompt) and documents connecting Claude and other AI clients to Cypress Cloud; these carried no weight here. Privacy policy, Cypress statement, 2023, Cloud MCP

Used Cypress App? Share your experience to help other readers choose.

WebdriverIO

Also consider · real Safari or existing drivers

$0MIT · OpenJS Foundation

Choose WebdriverIO when you must test branded Safari, Chrome, Firefox and Edge, or keep Selenium Grid, cloud-browser or Appium infrastructure, and can assemble and maintain more of the setup.

Reasons to choose it

  • Standard drivers for branded Chrome, Firefox, Edge and Safari
  • Mocha, Jasmine or Cucumber adapters
  • Optional DevTools interface and portable traces

What to watch

  • Browser session per spec file, not per test
  • Version 10 (October 5, 2026) removed older commands
  • Network mocking needs WebDriver BiDi; Safari not stated

WebdriverIO’s local runner runs Node.js tests through Mocha, Jasmine or Cucumber adapters, and its documentation lists standard drivers for Chrome, Firefox, Edge and Safari. That makes it the option here for teams that must automate the Safari their visitors use (Safari runs only on Apple platforms, so that means a Mac), or that already run Selenium Grid, a cloud browser provider or Appium for mobile apps. It is MIT-licensed and run by its community under the OpenJS Foundation. We verified that version 10.1.0 was released October 18, 2026. Overview, protocols, package record

The compromise is assembly and upkeep. Each spec file runs in its own worker and browser session for each browser configuration, so tests within one file share a session unless you reset state yourself. You choose a framework adapter, reporters and services, and each one is something else to keep updated. Version 10.0.0, released October 5, 2026, requires Node.js 22.19 or later, accepts only W3C WebDriver sessions, removes deprecated commands such as executeAsync and makes $ throw when a selector matches more than one element. Runner, v10 release, v10 migration guide

Read the full WebdriverIO analysis and sources

Network substitution needs WebDriver BiDi

WebdriverIO tries to start local sessions with the WebDriver BiDi protocol by default. Its mocking guide says the mock command “requires support for WebDriver Bidi”, which is “usually the case when running tests locally in a Chromium based browser or on Firefox”, and that a cloud provider must support BiDi too. The guide does not mention Safari, so check before designing Safari tests around replaced responses. In September the same guide named a Chrome DevTools Protocol requirement instead. Protocols, mocks and spies

Several browsers in one test, and visual debugging as a service

WebdriverIO’s multiremote mode runs several named browser sessions in one test, which can represent two users; version 10 renamed its APIs. Its optional DevTools service opens a local browser interface with the test tree, a browser preview, a command timeline with a screenshot after each command, network and console logs, single-test reruns and a portable trace.zip for offline replay. On browsers without a Chrome DevTools push path, its video and filmstrip fall back to taking screenshots over WebDriver, which adds round trips. We could not establish a minimum release or equal feature coverage across browsers. v10 migration guide, DevTools, DevTools limitations

Retries, reports and logged secrets depend on configuration

Retrying a single test or hook does not rerun the setup hook that accompanies it, while retrying a whole spec file starts a new browser instance and repeats setup. A repeat can therefore begin in a different state depending on the mechanism you choose. The JUnit reporter writes one XML file per spec file, giving you results to keep without a hosted dashboard. The security guide warns that masking has limits: cloud providers may log through a proxy that bypasses it, and reading or setting a masked field later can expose the value. Retries, JUnit reporter, security

The project’s one 2026 GitHub security advisory concerned an optional provider service: a command-injection flaw in @wdio/browserstack-service up to 9.23.2, fixed in 9.24.0 in May 2026 (CVE-2026-25244). Keep optional services on the same update schedule as the core packages. The OpenJS Foundation privacy policy covers its sites and services, not every plugin or remote provider. Advisory, OpenJS privacy

Used WebdriverIO? Share your experience to help other readers choose.

Which product are we missing?

Know a product that belongs in this article? Tell us what it does differently and why it deserves a closer look. Newcomers welcome.

Readers and product teams can suggest candidates. Suggestions inform our research; they do not guarantee inclusion or a recommendation.

Suggest a product

Opens an email draft with this article attached as a link. Nothing is sent until you send it.

No email app? Send the product name, website, and this article’s title to admin@practicalpicks.guide.

What people are saying

Public discussion is a check on project claims, not a measured reliability score. On npm, in the week of October 12 to 18, 2026, @playwright/test was downloaded about 37.7 million times, cypress about 2.8 million and webdriverio about 1.3 million; on October 19 their main GitHub repositories had about 98,000, 51,000 and 9,800 stars. Downloads count installs, including CI machines and other packages that depend on them. They indicate how large a community you can draw on for answers and examples, not quality. npm download counts, Playwright, Cypress, WebdriverIO

A Playwright user reported WebKit crashes in a specific Docker, GitLab and Kubernetes setup using version 1.43.1 in April 2024. The issue is closed, but we did not establish its detailed resolution, and it is not a present failure-rate estimate. Issue 30428

In 2023, Currents described disruption from Cypress blocking its integration. Cypress acknowledged targeted blocking and disputed the competitor’s account of the reasons; both have commercial interests. The practical lesson is to verify a proposed third-party dashboard integration rather than assume the runner’s open-source license guarantees it. Both posts were read September 7, 2026. Currents account, Cypress response

An April 2025 discussion of WebdriverIO, read September 7, 2026, includes contrary reports about community support and configuration effort, mostly around native mobile testing. Those experiences do not settle browser testing suitability or support response times; we used them to challenge assumptions about integration effort, not to score the framework. Discussion

How we researched

This is public-source research. We did not install any package, run any test, open any vendor account or upload any code. No project supplied access, payment or input, and we use no affiliate links. We first compared the three frameworks on September 7, 2026, and rechecked every material claim on October 19, 2026: releases and requirements, browser support, isolation, debugging and reporting features, prices, policies, security advisories and the Cypress incident. We read project documentation, changelogs and release posts, package records from the npm registry, the OSV vulnerability database and GitHub’s advisory records. Cypress Cloud’s separate terms did not load as readable text for our automated check.

We wrote this for a small team writing a new JavaScript or TypeScript end-to-end suite for its own web app, run locally and in CI without a required paid service. We fixed five criteria and their weights before comparing: outcome fit, including isolation, multiple users and browser coverage (30%); debugging and failure records available without a paid service (15%); setup and upkeep, including 2026 breaking changes (20%); total cost (15%); and risk and continuity, including security record, incidents and community size (20%). AI features carried no weight. On October 24, 2026 we re-scored cost under rule v1 on a three-year horizon; each fit runner’s cheapest suitable plan is $0, so every compared runner scores the maximum on cost and Playwright Test remains first at slopes 1, 1.5 and 2. Weighting capability alone, or debugging more heavily, did not change that. A hard requirement to test branded Safari is the condition that changes the choice, to WebdriverIO. Native mobile, component, load, visual-regression and AI-generated testing are outside this comparison. Selenium’s own bindings and Puppeteer provide automation without this kind of integrated test runner. TestCafe and Nightwatch are established open-source runners we priced at $0 for the same scenario; we did not fully score them because they are unlikely to outrank the three compared here on outcome fit and debugging for a new suite.

We also searched for new entrants through npm release records, GitHub, Hacker News launch posts, lists of alternatives and the projects’ own documentation on October 19 and again on October 24, 2026. We screened a TypeScript browser-test client first published in September 2026, TesterArmy’s open-source e2e framework announced in October 2026, OpenCheck and other tools that execute tests through a language model, Bombadil’s generative browser-testing project, a browser-automation tool for AI agents first released in December 2025, a mobile-testing tool’s web mode and several older AI-driven runners. None met our Rising Star threshold. Biloba still runs only Chrome and does not support Windows. TesterArmy e2e is pre-1.0 and centres on cached AI agent steps. OpenCheck and the older AI runners depend on a model to run tests. Bombadil targets property-based exploration rather than an ordinary scripted suite. New AI features from Playwright, Cypress and WebdriverIO are features of established products, not new entrants. We will look again at the next review.

ToolPublic sourcesWhat remains uncertain
Playwright TestIntroduction, best practices, browsers, isolation and trace documentation, release notes, test agents page, license, npm record, OSV advisory, Microsoft privacy statement.How often WebKit-build results differ from branded Safari for ordinary apps; setup time, speed and flakiness for your suite.
Cypress AppArchitecture, browser, isolation, configuration and reporter documentation, changelog, pricing page, privacy policy, website terms, incident disclosure, npm record.Cloud contract terms for refunds and early cancellation; taxes; the final forensic outcome and scope of the August 2026 incident.
WebdriverIOOverview, protocol, runner, mocking, DevTools, retry, JUnit and security documentation, v10 release post and migration guide, GitHub advisory, license, npm record.Behavior of version 10 in wider use; DevTools minimum release and browser parity; network mocking on Safari.
Rising Star and cost passOctober 24 newcomer search; cost rule v1 three-year workload and five-product reference set in the research record; Cypress Cloud Team optional price from cypress.io/pricing (Grakrawl).TestCafe and Nightwatch fit verdicts reuse October 19 docs without a fresh fetch today.

None of this establishes setup effort, execution speed, flaky-test rates or defect detection for your application. Retrying an assertion, rerunning a failed test and resetting application data are separate operations. Next substantive review is due December 18, 2026, or sooner after a major release of any of the three. Found a changed release or behavior? Contact the editors.

Sources and evidence

Sources were checked October 19, 2026 unless another date is given. Project material establishes documented features and published releases, not behavior in your environment. Community posts are attributed opinions, not representative samples.

This is a researched comparison, not a product review. Method: evidence standard, editorial picks.