Flathub installs: the Linux count, and the third channel to disagree
Flathub is the main Flatpak distribution point for Linux desktops, and it publishes per-application
install totals as open JSON. Like homebrew-cask-analytics and unlike every user-agent panel, it
identifies a browser by package name — com.brave.Browser — so brave shipping a Chrome
user-agent to defeat fingerprinting costs it nothing here. This is the Linux package count growth
edge 9a asked for, read 2026-08-25 from the v2 stats endpoint.
Cumulative installs
| Browser | Flathub installs | ~last 30 days |
|---|---|---|
| firefox | 12,053,090 | 202,410 |
| Google Chrome | 11,993,069 | 188,831 |
| brave | 6,483,260 | 173,856 |
| Microsoft Edge | 3,098,444 | 40,585 |
| Chromium | 2,056,629 | 27,777 |
| LibreWolf | 1,895,037 | 46,016 |
| zen | 1,340,706 | 70,272 |
| opera | 1,121,687 | 15,471 |
| vivaldi | 750,313 | 19,751 |
What it settles
Brave is 2.1× Edge on Flathub. On homebrew-cask-analytics it was 1.8× Edge. Two package managers, two operating systems, two populations of user, and the same ordering with almost the same ratio — which is the first time this spoke has had a Brave figure corroborated by anything. The Homebrew number could be dismissed as a macOS-developer artefact. This one cannot be dismissed the same way, and it agrees.
What it unsettles further
The opera/vivaldi pair now has three instruments and three answers. dap-browser-share put Opera ahead ~1,160:1 on US federal web traffic; Homebrew put Vivaldi ahead 1.9:1; Flathub puts Opera ahead 1.5:1. A ranking that flips with the channel is not a small-sample problem — the channels are sampling different people, and none of them is sampling the population.
The Flathub table also carries a name the user-agent panels barely register: zen at 1.34M, above Opera, and third by recent rate behind only Firefox and Chrome among the non-incumbents. A two-year-old Firefox fork out-installing Opera on this channel is either a real Linux-enthusiast result or a statement about who uses Flatpak. Both readings are about the channel.
Four limits, and they are not the Homebrew four
- Installs, not people, and not retention. A total counts events since the app was listed.
- One channel on a fragmented platform. Debian
.debrepositories, Arch, Fedora RPM, Snap and Brave's own installer script are all outside this count, and on Linux that is most of the ground. Distributions also differ in whether Flathub is enabled out of the box, which moves a browser's number without any user changing their mind. - No opt-out bias, unlike Homebrew. This is a server-side count of downloads, so the privacy correlation that makes the Homebrew figure a floor does not apply. Different bias, not less.
- Cumulative totals hide age. Firefox has been listed far longer than zen; the 30-day column is the fairer comparison and it tells a different story about who is growing.
Related
brave · homebrew-cask-analytics · dap-browser-share · brave-play-store-listing · statcounter-browser-share · zen · opera · vivaldi · synthesis