BROWSER USE Products: - [Web Agents](https://browser-use.com/web-agents) - [Stealth Browsers](https://browser-use.com/stealth-browsers) - [Custom Models](https://browser-use.com/custom-models) - [Proxies](https://browser-use.com/proxies) - [Skill APIs](https://browser-use.com/skills) [Pricing](https://browser-use.com/pricing) [Blog](https://browser-use.com/posts) [Cloud Docs](https://docs.cloud.browser-use.com) [Open Source Docs](https://docs.browser-use.com) [GET STARTED](https://cloud.browser-use.com) [GITHUB](https://github.com/browser-use/browser-use) --- # We Stealth Benchmarked Every Major Cloud Browser Provider **Author:** Aitor Mato **Date:** 2026-03-21 > We built a stealth benchmark from real production data and tested every major cloud browser. --- In [our first stealth post](https://browser-use.com/posts/bot-detection), we explained why current "stealth" solutions are fundamentally broken and how we forked Chromium to build the best browser infra. **Bold claim. This post is the proof.** ![Stealth Benchmark: Success Rate by Browser Provider](https://browser-use.com/images/stealth/accuracy_by_browser_dark.png) ## Why stealth needs its own benchmark The bottleneck for agents is not their intelligence. On most websites users care about, agents get blocked by a CAPTCHA or an antibot challenge. We collected 300,000 security check events from production traffic. The data made it clear: **agent intelligence and browser stealth are orthogonal**. You need to measure them separately. ## Dataset distribution From those 300,000 security checks, we extracted the 71 most common and well-known websites where agents were getting blocked, and categorized them by antibot vendor: | Vendor | Sites | | --- | --- | | Cloudflare | 23 | | PerimeterX | 18 | | Datadome | 13 | | Akamai | 8 | | reCaptcha | 6 | | Others (incl. Kasada, Shape) | 3 | | Total | 71 | > If you think a site or vendor is missing, let us know. ## Judge criteria - **Simple tasks**: 3 steps per site (search, visit a page, read a link). No auth. If it fails, it's because the browser got blocked. - **LLM judge** (`gemini-2.5-flash`): only evaluates whether the agent was blocked, not whether it completed the task correctly. - **Page load failures count as blocks**, since some antibots prevent the page from loading entirely. ## Control experiments To confirm the benchmark measures stealth, we tested two baseline browsers on a datacenter IP (GitHub Actions runner): - **Headless Chromium**: scored **2%**. The most suspicious browser possible, fewer than 2 tasks passed per run. - **Headful Chromium**: scored **50%**. Same machine, just rendering a visible window. Headless detection is one of the easiest signals for antibots. This validates that the task set consists of high-security sites and that score improvements reflect real stealth capability. ## Results We ran [the benchmark](https://github.com/browser-use/benchmark) across every major cloud browser provider. Each provider was evaluated multiple times with the same agent and model (`bu-2-0`). The differences are large and consistent. **Browser Use Cloud leads at 81%**, followed by Anchor at 77% and Onkernel at 67%. The bottom half (Steel 47%, Browserbase 42%, Hyperbrowser 40%) is far behind. ![Stealth Benchmark: Success Rate by Antibot Vendor](https://browser-use.com/images/stealth/category_heatmap_dark.png) The vendor breakdown reveals where each provider is strong or weak. Browser Use Cloud's highest bypass rates are against Cloudflare (93%), Akamai (85%), and PerimeterX (81%). ## BrowserBench We also ran [BrowserBench](https://halluminate.ai/blog/browserbench) from Halluminate, a third-party benchmark with 296 tasks across antibot-protected sites. Their task set is not fully focused on high-security websites, so overall scores are higher and differences between providers are smaller. ![BrowserBench: Success Rate by Provider](https://browser-use.com/images/stealth/browserbench_success_rate_dark.png) Browser Use Cloud achieved the highest success rate of all providers tested: - **Browser Use Cloud**: **84.8%** - **Hyperbrowser**: **76.4%** - **Anchor**: **76.0%** - **Steel**: **73.3%** - **Browserbase**: **70.3%** ## Should we trust benchmarks? Not blindly. Proxy quality varies by the hour, sites update their configs, CAPTCHA difficulty fluctuates. But **benchmarks show who is actually putting in the work**. When one provider consistently leads across multiple independent benchmarks, that's not noise. 71 sites is a start, not an endpoint. We know vendors and CAPTCHA providers are missing. We'll keep expanding and re-running as the landscape changes. ## "Stealth" in the industry Let's cut the bullshit. Most cloud browser providers slap a residential proxy and a third-party CAPTCHA solver extension on a stock Chromium, put **"Advanced Stealth"** on the landing page, and call it a day. 🤡 Most major competitors report the same fingerprint: **headful Linux Chromium.** The movie ends as soon as major antibots stop being lenient and start blocking these obvious leaks. We covered this in detail in [our first stealth post](https://browser-use.com/posts/bot-detection). TL;DR: if your provider's user-agent starts with `Mozilla/5.0 (X11; Linux x86_64)`, run. What we do instead: - **Fully headless on latest major Chromium version**. The only provider that can pull it off - **macOS, Windows & Linux** with tens of thousands of real fingerprints - **In-house CAPTCHA solvers** and monitoring tools - **Own Chromium fork**, maintained and patched non-stop Stealth is not a feature for us. It's the **top priority**. ## The never-ending cat & mouse game A solution that works today might break tomorrow. Antibots evolve constantly and no one is immune. We don't pretend to sell a 100% success rate, that doesn't exist. Things break, but nobody works harder than us to keep up. When your business relies on stealth, the most important thing when choosing a provider is **trust in their expertise and work ethic**. At Browser Use, we lead, and we prove it. If you are blocked or tired of your current provider, [give us a try](https://cloud.browser-use.com/browsers) or reach out at [aitor@browser-use.com](mailto:aitor@browser-use.com). We'll be happy to unblock you and do our best to make it work. ## What's next Next post will be about CAPTCHAs. How we solve them, what we build in-house, and how it works under the hood. We'll also compare features and metrics across providers. Shoutout to [Alexander Yue](https://www.linkedin.com/in/alexander-yue/) for building and maintaining the benchmark. The stealth benchmark is open source at [github.com/browser-use/benchmark](https://github.com/browser-use/benchmark).