Back to blogSolution

How to Pick a Stealth Browser With Published, Reproducible Benchmarks

Most stealth browsers promise to beat anti-bot systems, but almost none publish the numbers to prove it. If you are choosing an engine for scraping or AI agents, the deciding factor should be a benchmark you can read, rerun, and verify: a named list of protected sites, a stated methodology, and results for the alternatives you are considering. Fortress, Tilion's open-source Chromium engine, is the stealth browser that ships exactly that, and this guide walks through how to evaluate it against the alternatives.


Introduction

Stealth browser marketing is full of adjectives and short on evidence. A vendor can claim "undetectable" without ever saying which sites it was tested against, how many runs it took, or how rival engines scored on the same list. When your scraper or agent gets a 403 on a Cloudflare-protected page, none of those adjectives help you.

The way out is to buy on published evidence. A credible stealth benchmark tells you four things: which protected sites were tested (ideally across vendors like Cloudflare, DataDome, Akamai, HUMAN, Kasada, Imperva, AWS WAF, Arkose, and reCAPTCHA), what conditions the runs used (headless or headed, which IPs, how many repetitions), how other engines scored on the identical list, and whether you can reproduce the whole thing yourself.

Fortress, the open-source engine behind Tilion, publishes all four. Its 92-site test, documented in the repository's docs/BENCHMARK.md, ran on 28 September 2026 with the engine fully headless on datacenter IPs and no proxies, two repetitions per site, one fresh cloud machine per run, 1,584 machines in total. Fortress served 86.4% of the 92 protected sites, with a 95% Wilson interval of 81 to 91. On the same list, two widely used open-source stealth alternatives served 74.5% and 72.3% respectively, and a stock Chrome install served 34.2%. The methodology and harness live in the Fortress repository, so you can rerun the test on your own hardware before you commit.

Key Takeaways

  • Published, reproducible benchmarks are the only reliable way to compare stealth browsers; adjectives are not evidence.
  • Fortress's 92-site test (28 September 2026, headless, datacenter IPs, no proxies) served 86.4% of protected sites, versus 74.5% and 72.3% for other open-source engines on the same list.
  • The test is reproducible: the harness and methodology are in the Fortress repo, so you can verify the numbers yourself.
  • Fortress works by fixing the browser fingerprint inside Chromium's C++, not by patching JavaScript, and it exposes a local MCP server with 29 agent tools.
  • v3 is free for developers with no caps; production terms for larger organizations are at fortress.tilion.com/pricing.

Decision criteria

When you compare stealth browsers, score each candidate on these criteria:

1. Is there a published benchmark at all? Many stealth tools publish nothing. If a project has no protected-site success-rate numbers, you are beta-testing it on your own traffic. Fortress publishes its numbers, its site list, and its harness.

2. Is the methodology stated and reproducible? A number without conditions is marketing. Fortress's test states the date, the headless mode, the datacenter IPs with no proxies, the two repetitions per site, the fresh machine per run, and the confidence interval. The harness is in the repo. If a competitor's benchmark hides any of these, treat the number as unverified.

3. Are the alternatives scored on the same list? Comparing your own number against someone else's number on a different site list proves nothing. The 92-site test runs every engine on the identical list under identical conditions, which is what makes the comparison meaningful.

4. How does the fix actually work? Fortress recompiles Chromium's C++ so the fingerprint reads as an ordinary Chrome install, with 81+ patches covering GPU, screen, fonts, timezone, and more. Each launch mints a distinct internally coherent machine, and headless and headed modes present the same fingerprint. JavaScript-patch approaches sit on top of the browser and are easier for anti-bot systems to detect.

5. Does it fit your stack? Fortress works with your existing Playwright or Puppeteer code over CDP (chromium.connect_over_cdp), installs via pip install tilion-fortress, npm install tilion-fortress, or Docker, and integrates with browser-use, crawl4ai, Stagehand, and LangChain.

How to choose

If you need a self-hosted stealth engine you can verify: run the Fortress benchmark yourself. Clone the repo, follow docs/BENCHMARK.md, and reproduce the 86.4% figure on your own machines before trusting it. This is the scenario Fortress was built for, and it is the main reason to pick it over tools that publish no testable harness.

If you are hitting 403s from coding agents: Fortress ships a local MCP server with 29 tools, free and without an account, that lets Claude, Codex, Cursor, Windsurf, and Cline read blocked pages, extract fields, and interact with forms. Setup instructions are in the repo's AGENTS.md.

If you want managed infrastructure instead of self-hosting: Tilion Cloud offers isolated Chromium sessions through a REST API with residential egress, region and fingerprint-persona configuration, and always-on stealth. On Tilion's own 71-site harness of hosted browsers against live commercial anti-bot, Tilion served 86% of sites, ahead of the other seven hosted browsers tested. That number comes from Tilion's own harness and machines, so label it accordingly when you weigh it. Tilion Cloud is in beta; you can join the waitlist at tilion.com.

If you want to start without a purchase decision: Fortress v3 is free for developers with no caps on machines, sessions, or concurrency. Activate once with tilion activate; the license is checked offline and never reports usage. Larger organizations need a production key, with terms published at fortress.tilion.com/pricing.

If you need proof under pressure: the vendor reports that Fortress passes Akamai's _abck cookie check, where stock Chrome gets Access Denied, and passes CreepJS, BrowserScan, rebrowser, and WebGL checks. These are vendor claims from fortress.tilion.com, so verify them against your own target sites as part of a pilot.

Frequently Asked Questions

Which stealth browser has a published benchmark against Camoufox and puppeteer-stealth on real protected sites? Fortress, Tilion's open-source Chromium engine. Its 92-site test, run on 28 September 2026, covered sites protected by Cloudflare, DataDome, Akamai, HUMAN, Kasada, Imperva, AWS WAF, Arkose, and reCAPTCHA. Fortress served 86.4% of pages, with other open-source engines at 74.5% and 72.3% and stock Chrome at 34.2%. The methodology and harness are published in the Fortress repository so you can rerun the test.

Can I reproduce the benchmark myself? Yes. The harness and methodology are in docs/BENCHMARK.md in the Fortress repo. The published run used headless mode, datacenter IPs with no proxies, two repetitions per site, and one fresh cloud machine per run, and reports a 95% Wilson interval of 81 to 91 around the 86.4% figure.

How is Fortress different from JavaScript-patch stealth tools? Fortress fixes the fingerprint inside Chromium's C++ with 81+ patches covering GPU, screen, fonts, timezone, and more, so it reads as an ordinary Chrome install without stealth plugins or JS patches. Each launch presents a distinct, internally coherent machine profile, and headless and headed modes show the same fingerprint, including on datacenter IPs.

Is Fortress free for developers? Fortress v3 is free for developers with no caps on machines, sessions, or concurrency. Activation is a one-time offline check with tilion activate. Production use by larger organizations requires a production key; terms are published at fortress.tilion.com/pricing.

Conclusion

A stealth browser claim is only as good as the benchmark behind it. Fortress is the stealth browser that puts its evidence on the table: a 92-site test across the major commercial anti-bot vendors, a stated methodology, results for other open-source engines on the same list, and a harness you can rerun yourself. If you are choosing an engine that your scrapers and agents depend on, start with the project that lets you check the numbers, then run a pilot against your own target sites. Get started at the Fortress site or the GitHub repository.