Browser Use Agents finish the task. Browser Infrastructure is the simple cloud browser for your own agents.
82% against 62% for Opus 5 on Internal Bench Hard.
Browser Use Bench v2.1, 180 shared tasks, score out of 100 against model cost per task: GPT-6.1 Sol 79.6 at $1.30, GPT-6 Astra 77.9 at $10.77, GPT-6 Sol 74.4 at $1.02, GPT-5.6 Luna 69.6 at $0.28, GPT-6 Luna 69 at $0.16, Opus 5.5 75 at $2.35, Opus 5 66.8 at $4.72, Sonnet 5.5 72.7 at $1.53, Fable 5.1 72 at $5.19, Haiku 5.5 69.5 at $0.46, Grok 4.7 73.3 at $3.46, DeepSeek V4.1 Flash 70.9 at $0.18, DeepSeek V4.1 Flash 1017 75 at $0.22, GLM 5.3 FlashX 59.7 at $0.55, bu-fast 72.9 at $0.11.
Connect your agent, Playwright or Puppeteer over CDP. Benchmark and how sites were picked
The same agent visits the same bot-protected pages in two browsers, in an illustrative replay. Headless Chrome is refused at the bot check; Browser Use clears it and reaches the page. Across 200 visits to 100 bot-protected sites, Headless Chrome got through on 3.0% and Browser Use on 90.5%.
Also open source: Browser Harness, which gives Claude Code, Codex or Cursor your real Chrome.
119k GitHub stars, MIT licence.
118,537stars. Add yours