Browser automation

Online-Mind2Web (Ohio State) runs 300 live tasks on 136 real sites. Browser Use Cloud leads the current aggregation. Playwright MCP is the free local default by a wide margin of installs. Hosted browser prices have fallen to roughly $0.02–0.12 per browser-hour.

Independent benchmark Category 6 of 11
Independent benchmarkSteel browser-agent leaderboard (leaderboard.steel.dev) · Oct 11, 2026

Online-Mind2Web task success

  1. Browser Use Cloudbu-max 97.0%
  2. OpenAI computer useGPT-5.4 native 93.0%

Only the rows with an exact figure in our research are shown. On WebVoyager, a different benchmark, Skyvern 2.0 reports 85.85% (omxyz leaderboard).

Catalog only, not yet measuredVendor pricing pages, not individually re-verified · catalog compiled Nov 7, 2026

Hosted browser, entry price per browser-hour

  1. Browser Use Cloud $0.02
  2. Steel $0.08
  3. Hyperbrowser $0.10
  4. Browserbase ~$0.12

Shorter is cheaper. Browserbase's figure is a third-party estimate (proxlai, Oct 2026).

Tools

Tools with a benchmark score first, by score; the rest alphabetically. Links go to each vendor's own site.

Catalog only, not yet measuredvendor pages and secondary sources · compiled Nov 7, 2026
ToolWhat it doesEvidencePrice or terms
Browser Use Python browser-agent framework + cloud 97.0% Online-Mind2Web (Cloud)120k GitHub stars open source (MIT); Cloud from $0.02/browser-hour
Browserbase managed browsers; Stagehand native none yet free 1 browser-hour; $20/mo Developer
Chrome DevTools MCP drives a real Chrome over DevTools none yet free
Hyperbrowser hosted browsers with CAPTCHA solving none yet $0.10/browser-hour; 5,000 free credits
Playwright MCP accessibility-snapshot browser MCP 10.56M weekly npm downloads free (Apache-2.0)
Skyvern visual + DOM workflow agent 23k GitHub stars Hobby $29/mo, Pro $149/mo; self-host free
Stagehand act/extract/observe primitives on Playwright 26k GitHub stars open source (MIT)
Steel hosted browsers; open-source core none yet from $0.08/browser-hour; $30 credits

Caveats

  • Steel aggregates published results rather than re-running them, so stale self-reports carry through.
  • OSWorld figures in our research (Claude about 48%, Operator 44%) come from a comparison run by a vendor and are not shown as a ranking.

Other categories