Camoufox
Use a modified Firefox browser built for automation and anti-detection.
29 September 2026 · three fixed sites · 24 attempts · no proxy
Solo and three concurrent browsers
Patchright 1.63.0 and Camoufox 0.5.6, with two repetitions per mode on the same Web Scraping Dev, eBay and Indeed URLs. Compare content checks, capture time, worker-tree and aggregate peak PSS. Inspect every attempt and the exact guarded code; browser engines, filtering and fingerprints differ.
Inspect the resource and outcome comparison29 September 2026 · frozen public-30-v1 · no proxy
Camoufox on all 30 public targets
18 of 30 DOM captures met the unchanged target-specific checks. Inspect all 120 dated records across Camoufox, Patchright, Lightpanda and wreq, including failures, redirects, resource differences, hashes, guarded memory observations and exact code. Full captures remain private.
Compare all 30 targets across four tools29 September 2026 · this host · no proxy
One Camoufox run on Amazon laptop search
Camoufox 0.5.6 with browser 152.0.4-beta.31 received HTTP 503 on the exact URL used by the local Patchright attempt. The saved screenshot shows Amazon’s “Something went wrong” page, with no laptop product cards. Patchright returned HTTP 200 with 16 product cards minutes earlier. These two isolated attempts do not establish why Amazon returned different pages or which tool generally performs better.
29 September 2026 · same host and URL · no proxy
Three-worker Amazon search batches
In separate batches, Patchright returned product listings in all three attempts; Camoufox returned listings in two and an HTTP 503 error page in one. All three Patchright navigations overlapped, while only two Camoufox navigations overlapped because browser startup staggered them. The sampled batch PSS peaks were 1,274 and 1,651 MiB respectively. These small, sequential batches do not establish a general reliability, detection, or resource advantage.
Inspect all six attempts and raw capturesSeptember 2026 field test
Patchright and Camoufox on the same pages
One scrape per page plus three browsers at once. Patchright used less memory in this sample; the test does not prove a general bypass advantage. Inspect the exact statuses, screenshots, and script before using it for a real job.
Inspect the comparisonPublished test history
These May 2026 runs used one network without a proxy. Their pass counts describe those older attempts, not how the tool will perform today.
| Mode | Date | Version | Observed result | Network |
|---|---|---|---|---|
| camoufox | 2026-05-24 | 0.4.11 | 4/21 required-content checks | No proxy |
| camoufox tuned | 2026-05-25 | 0.6.0 | 11/21 required-content checks | No proxy |
| camoufox headed | 2026-05-24 | 0.4.11 | 3/21 required-content checks | No proxy |
Where it fits
High-value hard targets where a 50–65% unproxied success rate justifies slow, stateful, supervised browser sessions. Not for high-volume pipelines; pair with proxies to go further.
This guidance comes from the May 2026 review. Check a recent run before choosing for production.
What it can do
- Keep a browser session
- Render JavaScript
- Use a proxy
- Take screenshots
Capabilities are catalog notes, not outcomes from the benchmark above.
Project record
- Type
- Use a browser
- Ecosystem
- python
- Repo check
- 2026-05-23
- License
- MPL-2.0
Repository metadata is a dated snapshot, not a live health guarantee.
See how we validate pages, including what a pass does and does not mean.