browser-use

Pick Browser Use when the failure mode you're fighting is rejection, not logic. If your Playwright scripts time out behind a bot wall, or your agent burns twenty steps re-solving a CAPTCHA, the $0.02/hour cloud browser is the shortest path — point your existing code at it over CDP and nothing else in your stack changes. The second reason to pick it is cost-per-solved-task. The vendor publishes its own bench (BU Bench v2.1, 180 shared tasks) plotting score against dollars per task, and the spread is wide — cheap models land near $0.10–$0.16 per task while frontier models run past $10. That table is more useful for model selection than any leaderboard, though it's vendor-built and worth re-running on your own tasks. Pass when the site isn't defended. A plain HTTP request or a cheap VPS handles public pages for a fraction of the cost. Pass too on high-bandwidth jobs — pulling images, PDFs or large files through residential proxies at $5/GB quietly becomes the bill, and the direct/own-proxy route at $0.20/GB exists precisely for that, but you have to opt in. Watch the metering. Browser time rounds up to whole minutes with a one-minute minimum, so a hundred two-second health checks cost a hundred minutes. And new projects start at 10 concurrent sessions — you buy your way to 250 with a $1,000 top-up and 500 with $5,000, which suits teams already committed rather than anyone running a pilot. The closest alternative depends on what you're buying. Against Browserbase and Kernel, the decision usually turns on stealth reach and the open-source fallback: Browser Use is MIT licensed at 119k stars, so you can self-host the same library and keep the vendor cloud as an option rather than a lock-in. If you need the browser entirely inside your own network, the vendor cloud stealth