GitHub stars are the default way we judge a JS project's popularity. But stars measure who's watching a project, not who's shipping with it. So I pulled GitHub stars and npm weekly downloads for 14 JS frameworks to see how far apart "mindshare" and "actual usage" really are.
The gap was bigger than I expected.
The most-installed tools aren't the most-starred
Three anchors: Hono has 31k stars and 44.8M installs/week. TanStack Query, 49.8k stars and 58.5M installs/week. Next.js, 140k stars and 37.9M installs/week. So the two tools with a fraction of Next's stars both out-install it every single week.
Ranked by downloads-per-star — how much a tool gets used vs admired:
| Framework | downloads-per-star |
|---|---|
| Hono | 1,445x |
| TanStack Query | 1,175x |
| Next.js | 271x |
| Vue | 255x |
| Svelte | 55x |
| Remix | 24x |
downloads-per-star ranges ~60x across the set
That's the real headline. How much a tool gets used vs how much it gets admired spans from ~24x (Remix) to ~1,445x (Hono) — a 60x spread. Stars and actual usage are barely correlated at the top.
Why the ratio, not absolute installs?
npm download counts are noisy. CI pipelines re-install on every run, mirrors double-count, and now coding agents pull packages constantly. Absolute installs overstate real usage, and the inflation isn't even across tools.
The downloads-per-star ratio held up better as a signal: small drop-in libraries get added to package.json and shipped without anyone bothering to star them, while big-name frameworks pile up stars from people who never deploy with them.
You star what you admire. You install what you actually ship with.
The full 14
I wrote up all 14 frameworks with the method and raw numbers here: state of framework tools 2026.