BROWSER USE

- Browser Use Agents: give Browser Use a task and receive completed work. API V4 is current for new integrations.
- Browser Infrastructure: connect your agent or automation to managed browsers through SDK, REST, or CDP. Starts at $0.02/browser-hour.
- Developer tools: Open Source, Browser Harness, SDK, and MCP support the two products above.

[Developer Index](https://browser-use.com/index.md)
[Product Map](https://browser-use.com/llms.txt)
[Full Product Context](https://browser-use.com/llms-full.txt)
[Pricing](https://browser-use.com/pricing.md)
[Cloud Docs](https://docs.browser-use.com/cloud/quickstart)
[Open Source Docs](https://docs.browser-use.com/open-source/introduction)

---

# Browser Use alternatives (2026): what to use instead, and when

> The main Browser Use alternatives are Playwright for fixed scripts and CI, Stagehand for single AI steps in your own code, Skyvern for no-code form workflows, Firecrawl for bulk extraction, Playwright MCP, Chrome DevTools MCP or agent-browser for coding agents, and Kernel or Browserbase for hosted browsers. In our matched Codex eval none of those coding-agent tools scored significantly differently from our Browser Harness, which used the fewest tokens. Browser Use is built for the case where one prompt should run a whole multi-step task, logins included.

[HTML page](https://browser-use.com/browser-use-alternatives) · [Cloud browsers compared](https://browser-use.com/cloud-browsers.md) · [Agent benchmark](https://browser-use.com/benchmarks/agents) · [Stealth benchmark](https://browser-use.com/benchmarks/browsers) · [Coding-agent eval data](https://browser-use.com/benchmarks/coding-agent-browser-tools.json)

Names used here: **browser-use** is the open-source Python agent library (MIT). **Browser Use Cloud** is the hosted product (agents API and stealth browsers); it is not open source and cannot be self-hosted. **Browser Harness** (MIT) gives coding agents your real Chrome.

## Pick by job

| Job | Alternatives | Browser Use |
|---|---|---|
| An agent that runs the whole task | Skyvern, Browserbase Agents, OpenAI Agents API computer use; Stagehand step by step | browser-use (MIT), Browser Use Cloud agents |
| Scripted steps you own | Playwright, Puppeteer | Cloud browsers run your script over CDP |
| A browser for a coding agent | Playwright MCP, Chrome DevTools MCP, agent-browser | Browser Harness (MIT) |
| Hosted browsers | Browserbase, Kernel, Steel, Browserless, Anchor | Browser Use Cloud browsers |
| Bulk extraction API | Firecrawl | No product for this |
| Whole-desktop control | Claude and OpenAI computer use | No product for this |

## Pick by what you need

Rows reflect common reasons people give for switching. Where another tool fits better, the row says so. Facts other tools state about themselves come from their own pages, checked October 15, 2026; latency is from BrowserArena and stealth from our Stealth Bench V2.

| If you need | Use instead | Why | What Browser Use has for it |
|---|---|---|---|
| Fixed steps or CI tests | Playwright | Code you own, the same result every run, no model call per step. JavaScript and TypeScript, Python, Java and .NET. | Browser Use Cloud browsers run your existing Chromium Playwright script over CDP: swap `launch()` for `connectOverCDP()`. $0.02 per browser-hour. [Remote browser for Playwright](https://browser-use.com/playwright.md) |
| AI on single steps inside your own script | Stagehand | act(), extract() and observe() inside a script you write, in TypeScript, Python and Go. On Browserbase browsers, caching is on by default: once an identical call passes a hit threshold, it comes back without spending tokens. | The browser-use library is Python. From TypeScript, browser-use-sdk calls the hosted agent: send a task, get JSON back. [Web agent API](https://browser-use.com/web-agent-api.md) |
| A browser for Claude Code, Codex or Cursor | Playwright MCP, Chrome DevTools MCP, agent-browser | A tool the coding agent drives, not a second agent loop. None scored significantly differently from Browser Harness in our matched Codex eval. | Browser Harness (MIT). On 100 of the 200 BU Bench V2.1 tasks: Browser Harness 60.4, Playwright MCP 60.0, Chrome DevTools MCP 61.0, agent-browser 56.3, no significant difference. Median input tokens per task: 2.1M against 3.3M to 6.7M. [The matched eval](https://browser-use.com/agent-browser) |
| Lower model cost per task | Playwright for known steps; Stagehand with Browserbase caching | Every agent step is a model call, so cost grows with the number of steps. Code that calls no model costs nothing per run. | Browser Use's own agent ranks #6 of 19 configs on our BU Bench V2.1: 72.9 at 11¢ a task in model cost, level with Claude Sonnet 5.5 high (72.7, $1.53). Top: GPT-6.1 Sol medium, 79.6 at $1.30. [Agent benchmark](https://browser-use.com/benchmarks/agents) |
| Form-heavy portal workflows, built without code | Skyvern | Vision LLMs on top of Playwright, with a no-code workflow builder. Free, $29 or $149/month, Enterprise custom (2FA / TOTP support and 1Password integration on Pro). AGPL-3.0. | The same login handling (TOTP, emailed codes, 1Password, saved profiles) pay as you go with no plan, from code or one prompt. A saved profile keeps the site from asking again while the session stays valid. [Login automation](https://browser-use.com/login-automation.md) |
| Hosted browsers for your own agent | Kernel, Browserbase | Of the 7 providers we compare, Kernel is fastest on BrowserArena: 318 ms median to create, connect, load a page and release a session (run of October 14, 2026), against 493 ms for Browser Use Cloud. Browserbase if Stagehand is central to your stack. | Browser Use Cloud: $0.02 per browser-hour, the lowest of 7 providers compared. It got through on 90.5% of visits on our [Stealth Bench V2](https://browser-use.com/benchmarks/browsers), on sites picked where it historically got through. [7 cloud browsers compared](https://browser-use.com/cloud-browsers.md) |
| Browsers you host yourself | Steel, Kernel, Browserless | [Steel](https://browser-use.com/alternatives/steel.md)'s browser server, steel-browser, and the browser images behind [Kernel](https://browser-use.com/alternatives/kernel.md)'s managed service, kernel-images (both Apache-2.0), run in your own Docker. [Browserless](https://browser-use.com/alternatives/browserless.md) runs in Docker or Kubernetes and is free for non-commercial use; commercial self-hosting needs its paid license. | Browser Use Cloud is hosted only, with an EU region. The browser-use library (MIT) runs on your own machine. [Self-hosting compared](https://browser-use.com/cloud-browsers.md) |
| Bulk extraction to markdown or JSON | Firecrawl | Scrape, crawl and search APIs that return markdown or JSON. Its agent endpoint searches, navigates, and gathers data, with 5 free runs a day; /interact and Browser Sandbox also click, fill forms and log in. | Browser Use when one prompt should run a whole multi-step task, not just return page data. [Browser Use vs Firecrawl](https://browser-use.com/posts/browser-use-vs-firecrawl) |
| Control of a whole desktop | Claude or OpenAI computer use | Screenshots plus mouse and keyboard on any app. Claude computer use runs on a machine you provide; OpenAI also offers a hosted browser in its Agents API. | Browser only. Browser Use reads the page structure, not just pixels, and hosts the browser for you. [Claude in your browser](https://browser-use.com/claude-computer-use) |
| A personal assistant, no code | Perplexity Comet, Claude in Chrome, ChatGPT | A person delegating one-off tasks in a chat or a browser. The Comet browser is a free download; some assistant features need a paid Perplexity plan. | Not a consumer app. Browser Use is for developers: an API, your choice of model, many runs in parallel. |

## Against scripts and agent SDKs

### Browser Use vs Playwright

Playwright wins when you know the steps: deterministic code in JavaScript and TypeScript, Python, Java and .NET, with no model call and no model bill. Browser Use wins when pages change or the task is written in plain language. They combine: Browser Use Cloud browsers take any Chromium Playwright script over CDP at $0.02 per browser-hour. [How to connect](https://browser-use.com/playwright.md) · [Why the library moved from Playwright to CDP](https://browser-use.com/posts/playwright-to-cdp)

### Browser Use vs Stagehand

Stagehand (MIT, by Browserbase) puts act(), extract() and observe() inside a script you write, in TypeScript, Python and Go, so you keep control and checks between steps. Browser Use plans and runs the whole task from one prompt, as a Python library or a hosted API with Python and TypeScript SDKs. No shared public benchmark covers both, so we quote none. Pick Stagehand to keep control between steps; pick Browser Use to hand over the whole task. [Stagehand](https://github.com/browserbase/stagehand)

### Browser Use vs Skyvern

Skyvern (AGPL-3.0) uses vision LLMs on top of Playwright and has a no-code workflow builder; plans are free, then $29 or $149/month, Enterprise custom, with 2FA / TOTP support and 1Password integration on Pro. Pick it for repetitive form workflows built by non-developers. Pick Browser Use for an MIT library or pay-as-you-go cloud with no monthly plan. [Skyvern pricing](https://www.skyvern.com/pricing)

## Against browser tools for coding agents

We ran a matched eval: 100 of the 200 public BU Bench V2.1 tasks, the same agent (Codex with GPT-6 Luna), the same fresh Browser Use Cloud browser and the same judge; only the browser tool differs. One run per task. Every paired difference against Browser Harness has a 95% interval that includes zero. Claude Code and Cursor were not measured.

| Tool | Mean score (0-100) | Median input tokens per task |
|---|---:|---:|
| Chrome DevTools MCP | 61.0 | 6.7M |
| Browser Harness | 60.4 | 2.1M |
| Playwright MCP | 60.0 | 4.4M |
| agent-browser | 56.3 | 3.3M |

### Browser Use vs agent-browser

agent-browser (Vercel, Rust CLI, Apache-2.0) is a browser tool for coding agents; Browser Use's tool for that job is Browser Harness, not the browser-use library. On 100 of the 200 BU Bench V2.1 tasks the difference was not significant: agent-browser 56.3, Browser Harness 60.4 (paired difference 4 points, 95% interval -1.8 to 10, so inconclusive), at 3.3M vs 2.1M median input tokens per task. Both ran on fresh Browser Use Cloud browsers. [Side by side](https://browser-use.com/agent-browser)

### Browser Use vs Playwright MCP

60.0 against 60.4, no significant difference, at 4.4M median input tokens against 2.1M. Pick Playwright MCP if you want Playwright's accessibility snapshots and already use MCP; its `--extension` mode also attaches to your logged-in Chrome. Pick Browser Harness for fewer tokens per task in our eval. [Side by side](https://browser-use.com/playwright-mcp)

### Browser Use vs Chrome DevTools MCP

61.0 against 60.4, no significant difference, at 6.7M median input tokens against 2.1M. Pick it for performance traces and DevTools debugging; Browser Harness for fewer tokens. [Side by side](https://browser-use.com/chrome-devtools-mcp)

## Against cloud browser vendors

### Browser Use vs Browserbase

Browserbase is $0.10 to $0.12 per browser-hour after 100 or 500 hours included in $20 or $99/month plans; Browser Use Cloud is $0.02, no subscription. On our [Stealth Bench V2](https://browser-use.com/benchmarks/browsers), every provider with its own proxy, stealth mode and CAPTCHA solver on (Browser Use Cloud 84% with its solver off), Browser Use Cloud got through on 90.5% of visits and Browserbase on 78.5% (standard mode; its Enterprise-only Verified mode was not tested); the sites were picked because Browser Use Cloud historically got through where stock Chrome with a US proxy did not, so they are hard on purpose and not a balanced sample of the web or of anti-bot vendors. Pick Browserbase if Stagehand is central to your stack. Browserbase also sells hosted Agents ("Scale fully managed browser agents with simple prompts"), which compete with Browser Use Cloud agents. [Browserbase pricing](https://www.browserbase.com/pricing) · [Browserbase alternatives](https://browser-use.com/alternatives/browserbase.md) · [Browser Use vs Browserbase](https://browser-use.com/posts/browser-use-vs-browserbase) · [All 7 compared](https://browser-use.com/cloud-browsers.md)

### Browser Use vs Kernel

Of the 7 providers we compare, Kernel is fastest on BrowserArena: 318 ms median to create, connect, load a page and release a session (run of October 14, 2026), against 493 ms for Browser Use Cloud. Kernel is $0.06 to $0.48 per browser-hour (headless to headful, billed per GB-second; plans $0 to $200/month); its browser image is open source (Apache-2.0) and runs in Docker. Browser Use Cloud is $0.02 with no subscription and got through on 90.5% of visits on our Stealth Bench V2 against 80.5% for Kernel; the sites were picked because Browser Use Cloud historically got through where stock Chrome with a US proxy did not, so they are hard on purpose and not a balanced sample of the web or of anti-bot vendors. [Kernel alternatives](https://browser-use.com/alternatives/kernel.md)

## Against extraction APIs and computer use

### Browser Use vs Firecrawl

Firecrawl (AGPL-3.0) is built for extraction: scrape, crawl and search APIs return markdown or JSON at scale, and its agent endpoint searches, navigates, and gathers data, with 5 free runs a day. Its scrape API can run up to 50 scripted actions first, and its [/interact](https://docs.firecrawl.dev/features/interact) endpoint and [Browser Sandbox](https://docs.firecrawl.dev/features/browser) click buttons, fill forms, scrape behind a login and reuse an authenticated session from a named profile. Pick Firecrawl when the output is data from many pages. Pick Browser Use when one prompt should finish a multi-step task on a site. [Browser Use vs Firecrawl](https://browser-use.com/posts/browser-use-vs-firecrawl)

### Browser Use vs OpenAI computer use and ChatGPT

OpenAI offers two forms. The Responses API computer tool returns actions or code; you provide the environment and execute the model's requests. The Agents API runs the browser in an OpenAI-hosted environment. Both run OpenAI models. The agent inside ChatGPT is for a person in a chat. Browser Use gives you a choice of model, an open-source MIT library, and stealth browsers at $0.02 per browser-hour. [Computer use guide](https://developers.openai.com/api/docs/guides/tools-computer-use) · [Agents API computer use](https://developers.openai.com/api/docs/guides/agents-api/tools/computer-use)

### Browser Use vs Claude computer use

Claude computer use controls a whole desktop from screenshots, roughly 1,000 to 1,800 input tokens each; your application runs every call in an environment you control. Pick it for desktop apps. For browser-only tasks, Browser Use reads the page structure and supplies the hosted browser. [Anthropic docs](https://platform.claude.com/docs/en/agents-and-tools/tool-use/computer-use-tool) · [Claude in your browser](https://browser-use.com/claude-computer-use)

## Questions people ask

**What is the best alternative to Browser Use?** The main Browser Use alternatives are Playwright for fixed scripts and CI, Stagehand for single AI steps in your own code, Skyvern for no-code form workflows, Firecrawl for bulk extraction, Playwright MCP, Chrome DevTools MCP or agent-browser for coding agents, and Kernel or Browserbase for hosted browsers. In our matched Codex eval none of those coding-agent tools scored significantly differently from our Browser Harness, which used the fewest tokens. Browser Use is built for the case where one prompt should run a whole multi-step task, logins included.

**Is Browser Use open source?** The browser-use library is open source under MIT: a Python agent that drives a browser on your own machine. Browser Use Cloud is a separate hosted product, not open source and not self-hostable: agents through an API, and stealth browsers at $0.02 per browser-hour with no subscription. Browser Harness, for coding agents, is open source under MIT.

**Browser Use vs Browser Harness: what is the difference?** Both are from Browser Use. The browser-use library is an agent: it runs its own model loop and finishes a task from one prompt. Browser Harness (MIT) has no agent loop; it gives Claude Code, Codex or Cursor your real Chrome over CDP so the coding agent drives it.

**Browser Use vs Playwright: which should I use?** Playwright when you know the steps: code you own, the same result every run, no model cost. Browser Use when the page or the task changes and a model has to decide each step. You can combine them: Browser Use Cloud browsers take any Chromium Playwright script over CDP at $0.02 per browser-hour.

**Browser Use vs Stagehand: what is the difference?** Stagehand, from Browserbase, adds act(), extract() and observe() to a script you write, in TypeScript, Python and Go. Browser Use runs the whole task from one prompt, as a Python library or as a hosted API with Python and TypeScript SDKs. Pick Stagehand to keep control between steps; pick Browser Use to hand over the whole task.

**What should I give Claude Code, Codex or Cursor to browse the web?** Any of Browser Harness, Playwright MCP, Chrome DevTools MCP or agent-browser; all four can attach to your logged-in Chrome. On 100 of the 200 BU Bench V2.1 tasks, with the same agent (Codex, GPT-6 Luna), browser and judge, they scored Browser Harness 60.4, Playwright MCP 60.0, Chrome DevTools MCP 61.0, agent-browser 56.3, no significant difference. Browser Harness used the fewest median input tokens per task in that eval, 2.1M against 3.3M to 6.7M.

**Can I self-host Browser Use?** The browser-use library (MIT), yes: it runs on your machine with your own model key and browser. Browser Use Cloud, no: it is hosted only, with an EU region. For self-hosted browser infrastructure, Steel's browser server (steel-browser) and Kernel's browser images (kernel-images; its hosted service stays managed) are open source and run in Docker; Browserless self-hosting needs a commercial license for commercial use.

**Is Browser Use expensive to run?** An agent pays for a model call on every step, so long tasks cost more than a script. Browser Use's own agent ranks #6 of 19 configs on our BU Bench V2.1: 72.9 at 11¢ a task in model cost, level with Claude Sonnet 5.5 high (72.7, $1.53). Top: GPT-6.1 Sol medium, 79.6 at $1.30. That is on 180 tasks, our agent in its own harness and the other models in BrowserCode. Browsers are $0.02 per browser-hour, no subscription. For steps you already know, Playwright costs nothing per run.

## Related

- [Cloud browsers compared](https://browser-use.com/cloud-browsers.md)
- [Browserbase alternatives](https://browser-use.com/alternatives/browserbase.md) · [Browserless alternatives](https://browser-use.com/alternatives/browserless.md) · [Steel alternatives](https://browser-use.com/alternatives/steel.md) · [Kernel alternatives](https://browser-use.com/alternatives/kernel.md)
- [Playwright MCP](https://browser-use.com/playwright-mcp) · [Chrome DevTools MCP](https://browser-use.com/chrome-devtools-mcp) · [agent-browser](https://browser-use.com/agent-browser)
- [Agent benchmark](https://browser-use.com/benchmarks/agents) · [Stealth benchmark](https://browser-use.com/benchmarks/browsers)
- [Security](https://browser-use.com/security.md) · [Browser Use Agents](https://browser-use.com/web-agents.md)

## Sources

- Matched coding-agent eval: [method and per-task scores](https://browser-use.com/benchmarks/coding-agent-browser-tools.json), [BU Bench V2.1 tasks](https://github.com/browser-use/benchmark)
- Agent scores and cost per task: [BU Bench V2.1 leaderboard](https://browser-use.com/benchmarks/agents)
- Cloud browser prices, stealth and BrowserArena latency: [Cloud browsers compared](https://browser-use.com/cloud-browsers.md), [BrowserArena](https://www.browserarena.ai)
- Playwright: [supported languages](https://playwright.dev/docs/languages), [repository](https://github.com/microsoft/playwright)
- Stagehand: [README](https://github.com/browserbase/stagehand), [caching](https://docs.stagehand.dev/v4/best-practices/caching)
- Skyvern: [README](https://github.com/Skyvern-AI/skyvern), [pricing](https://www.skyvern.com/pricing)
- Browserbase Agents: [pricing](https://www.browserbase.com/pricing)
- Firecrawl: [agent docs](https://docs.firecrawl.dev/features/agent), [interact](https://docs.firecrawl.dev/features/interact), [Browser Sandbox](https://docs.firecrawl.dev/features/browser), [scrape actions](https://docs.firecrawl.dev/advanced-scraping-guide), [pricing](https://www.firecrawl.dev/pricing), [repository](https://github.com/firecrawl/firecrawl)
- agent-browser: [repository](https://github.com/vercel-labs/agent-browser); Playwright MCP: [repository](https://github.com/microsoft/playwright-mcp); Chrome DevTools MCP: [repository](https://github.com/ChromeDevTools/chrome-devtools-mcp)
- OpenAI: [computer use guide](https://developers.openai.com/api/docs/guides/tools-computer-use), [Agents API computer use](https://developers.openai.com/api/docs/guides/agents-api/tools/computer-use); Anthropic: [computer use tool](https://platform.claude.com/docs/en/agents-and-tools/tool-use/computer-use-tool)
- Perplexity Comet: [free for everyone (October 2025)](https://www.perplexity.ai/hub/blog/comet-is-now-available-to-everyone-worldwide)
- browser-use-sdk: [npm](https://www.npmjs.com/package/browser-use-sdk), [PyPI](https://pypi.org/project/browser-use-sdk/)