Browser Use

Browser Use Agents

Forms are the boring part. Hand them over.

Give it the URL and the data. It finds the fields by their labels, clicks through the wizard, solves the CAPTCHA, and returns the confirmation as JSON. A thousand of them, in parallel.

Run your first formMore recorded runs
Booking a driving test: pick a date, solve the CAPTCHA, fill the form, confirm. 20 seconds, unedited, from the README.
82%
of our hardest tasks solved
17¢
per solved task on that benchmark
1,000+
forms at once on enterprise terms; 10 to start

One form, start to finish

  1. 1

    You describe it

    The URL and the values in one sentence, or a template filled from a row of your spreadsheet. Add an output schema for the confirmation fields you want back.
  2. 2

    It fills and submits

    Labels and roles tell it which field is which. Dropdowns, dates and uploads are ordinary actions. It pauses where you told it to if a person should look first.
  3. 3

    You get the receipt

    Confirmation numbers, screenshots and downloaded PDFs come back in the result and the workspace. Failures come back with the step and the page state. Behind a login? See login automation.

One run per row.

Runs are independent, so a thousand applications are a thousand browsers working at once. Concurrency starts at 10 and rises with spend to 1,000. Each run bills model tokens plus browser time at $0.02 per hour, and $5/GB if it uses the residential proxies.
from browser_use_sdk.v4 import BrowserUse applicant = {"name": "Ada Lovelace", "email": "ada@example.com", "plan": "Business"} with BrowserUse() as client:    run = client.runs.create(        f"Open https://example.com/signup and submit the form with {applicant}. "        "Stop before any payment step and report the confirmation number."    )    print(client.runs.wait_for_completion(run.id).result)
fill.pyAPI V4 · Python and TypeScript SDK

Measured on the standard live-web benchmark.

Online-Mind2Web runs on real websites: shopping, finance, travel, government forms. All 300 tasks, none removed.

Browser Use Cloud (v4): 98% accuracy. 12 points ahead of ABP + Opus 4.6, the next best.

  • Browser Use Cloud (v4)98%
  • ABP + Opus 4.686%
  • TinyFish81%
  • Navigator78%
  • Gemini CUA69%
  • Stagehand (Gemini 2.5 CU)65%
  • OpenAI Operator61%
  • Sonnet 4.0 CU61%
  • Stagehand (Sonnet 4.5)55%

Accuracy · longer is better · scale 0–100%

OnlineMind2Web · updated 2026-03-25 · all benchmarks
Every agent benchmark

Questions people ask

How does it know which field is which?

It reads the labels and roles on the page, the same things a person reads, so it fills by meaning instead of by CSS selector. When the portal moves a field, the same task still works.

Multi-page wizards, dropdowns, uploads?

Yes. Custom dropdowns, date pickers, multi-step forms, uploads from the workspace and downloads of confirmations are ordinary browser actions to it.

What about the CAPTCHA on the submit button?

Solved inside the cloud browser at no extra charge, like in the recording above. If one refuses to go, a person can take over through the live view and the run carries on.

How do I keep it from doing something I cannot undo?

Say so in the task, for example stop before payment. The run pauses at that step and waits in the live view. Domain allow lists keep it on the sites you name.

Can it fill forms behind a login?

Yes. Start from a saved profile, sync the cookies from your own Chrome, or bind a 1Password item so the password is typed without the model ever seeing it.

What does a form cost?

Model tokens plus browser time at $0.02 per hour. On our hardest benchmark the agent solves 82% of tasks at 17¢ per solved task; a short form costs less than that.