Browser Use is the most complete open-source browser agent: you hand it a goal and it drives the browser until the job is done. In 2026 it added a hosted cloud with the lowest published browser-hour rate in the category and its own action model.
Key Takeaways
- The whole agent, open source. — The browser-use Python library (MIT) takes a task in plain language and drives a browser until it is done, with any LLM, including local models through Ollama.
- The cheapest published browser time. — Browser Use Cloud bills browsers at $0.02 per browser-hour, agent steps from $0.006 and $0.01 per task, with no subscription required.
- Its own model. — BU 2.0, released January 27, 2026, is a model trained to pick the next browser action. Browser Use's own benchmark puts it at 63.3% task success, behind the strongest frontier model it tested.
What does Browser Use do?
The browser-use library is a Python package (3.11 or later, MIT license; version 0.13.14 shipped on October 26, 2026). It reads the page, asks the model for the next action, executes it and repeats until the task is done. You can run it on your own machine with any LLM. Browser Use was founded by Magnus Müller and Gregor Žunič and raised a $17 million seed round led by Felicis in March 2025.
BU 2.0, released January 27, 2026, is the company's own model for choosing browser actions. On BU Bench V1, Browser Use's 100-task benchmark drawn from WebBench, Mind2Web 2, GAIA, BrowseComp and custom tasks, it scored 63.3% through the open-source library. That is a vendor benchmark; the same chart shows a stronger frontier model at 80%. The cloud adds hosted browsers, proxies and an API that runs the agent for you.
Scorecard: 19/20
Four criteria, 1 to 5 each: how much of the agent loop the product supplies, the hosted browser infrastructure, openness (license, self-hosting, model choice), and published cost. These are editorial assessments of documentation and pricing pages, not benchmark runs. They match the Browser Automation for AI Agents guide.
| Criterion | Score | Why |
|---|---|---|
| Agent | 5/5 | A full autonomous loop from a plain-language task, plus a browser-specialized model. |
| Infrastructure | 4/5 | Hosted browsers, residential proxies, concurrency up to 1,000+ sessions, SOC 2 Type II, HIPAA mode on Enterprise. |
| Openness | 5/5 | MIT library, runs locally, works with any LLM provider or local model. |
| Cost | 5/5 | $0.02 per browser-hour, pay as you go, credits that do not expire. |
How much does Browser Use cost?
| Item | Price |
|---|---|
| Library | Free (MIT); you pay your LLM provider |
| Cloud browser | $0.02 per browser-hour, billed per minute |
| Agent step / task start | From $0.006 per step; $0.01 per task |
| Managed residential proxy | $5 per GB |
| Concurrency | 10 sessions free, rising to 1,000 as total spend reaches $25,000; Enterprise custom |
As published on browser-use.com, checked November 3, 2026. New subscriptions are no longer sold; credits are prepaid and do not expire. Enterprise adds SOC 2 Type II, a HIPAA mode, zero data retention and EU data residency on request.
Who it fits, and who should look elsewhere
- Fits: Python teams that want an agent to handle open-ended web tasks, and anyone who wants to start locally and move to hosted browsers without changing libraries.
- Look elsewhere: if the workflow is known and must be repeatable. Scripted steps with AI fallbacks in Stagehand cost fewer tokens and fail more predictably. For long-lived logged-in sessions and session replay, Browserbase has the deeper platform.
Browser Use alternatives
- Stagehand: code-first automation with AI primitives, MIT, in TypeScript, Python and more.
- Browserbase: managed browsers with contexts, CAPTCHA solving and a one-call Agents API.
- Steel, Hyperbrowser and Playwright MCP: scored in the full guide.
Related questions
- Which browser infrastructure should developers use for AI agents?
- Browser Use, Browserbase or Stagehand: which fits?
- When should you use an AI agent instead of scripted automation? (partner site CTAIO)
- How do you evaluate an AI agent before production? (partner site CTAIO)
- How does prompt injection work? (partner site CTAIO)
Sources
- Browser Use, Pricing (checked Nov 3, 2026).
- PyPI, browser-use 0.13.14 (Oct 26, 2026).
- Browser Use, BU 2.0 model card (Jan 27, 2026).
- ActuIA, Browser Use raises $17 million (Mar 26, 2025).
Is Browser Use free?
The browser-use library is free and MIT-licensed; you pay only for the LLM you connect. Browser Use Cloud is pay as you go: $0.02 per browser-hour, agent steps from $0.006, $0.01 per task and $5 per GB of managed residential proxy, as published in November 2026. New eligible sign-ups get a one-time $1 credit.
What is Browser Use used for?
Building AI agents that complete tasks in a real browser: filling forms, extracting data from sites without an API, testing flows and running research. You give it a goal in plain language and it decides the clicks and typing.
Which LLMs work with Browser Use?
Any major provider through wrappers for OpenAI, Anthropic and Google, local models through Ollama, and Browser Use's own BU 2.0 model, which is tuned for choosing browser actions. In the cloud, bring-your-own-key usage carries a 0.2x orchestration fee on the provider's token rates.
Browser Use or Stagehand?
Browser Use if you want the agent to decide every step from a goal. Stagehand if you want to write the workflow in code and use AI only for the steps that are hard to script. See our three-way comparison with Browserbase for the details.
Ready to Find the Right AI Tools?
Browse our data-driven rankings to find the best AI tools for your team.