Menu

Reproducible browser benchmarks

Browser automation and browser agents, measured — with the scripts.

BrowseBench runs the same tasks through browser automation libraries and AI browser agents and publishes success rates, timings, raw data and the code. If we haven't run something, we say so.

Latest result

Round 0: Playwright vs Puppeteer

10 tasks × 10 runs each, same Chromium binary, 2026-11-21.

PlaywrightPuppeteer
Runs passed100/100100/100
Sum of per-task medians19,804 ms19,355 ms

AI browser agents

Round 1: Claude Code as a browser agent — Claude Code (headless) + Playwright MCP with three Claude models on 10 tasks with checkable answers, against a scripted baseline. Next, Round 2 adds Browser Use, Stagehand and Skyvern. All rounds · scripts on GitHub

Tools we track

11 libraries, agent frameworks and browser services, with licence and activity data from GitHub. See the list, read browser automation tools compared with our data or jump to a comparison: Playwright vs Puppeteer · Browser Use vs Stagehand.