Frontend Frameworks
Figures come from the npm registry, the npm downloads API, jsDelivr and GitHub on 2026-08-06.
Three measurement caveats, and they are large enough to invert rankings. First, npm counts inflated unevenly through 2026 - Solid grew 6.8x, React 4.3x, Angular 1.5x over fifteen months, and nothing in frontend gained four times its real users in that window. Second, npm systematically under-counts script-tag libraries: on jsDelivr, jQuery serves roughly 6x React’s traffic and Alpine roughly 2x, the exact inverse of the npm ranking. Third, GitHub’s open-issue count includes pull requests.
Summary
React 19.2.8 is dominant and no longer Meta’s. The React Foundation launched under the Linux
Foundation on 2026-02-24, and github.com/facebook/react now redirects to github.com/react/react.
Vue 3.5.41 has shipped patches only for 23 months. Vapor mode is at release candidate, not
stable, whatever blog posts say. One person writes ~31% of its commits.
Angular 22.1.0 has by far the healthiest maintenance here - 4,044 commits from 232 contributors -
and has quietly completed the most radical architecture change of any framework, from zone-patching to
signals, without a rewrite-your-app moment.
Svelte 5.56.8 ships continuously - 56 minors in 22 months, no Svelte 6 - with two Vercel-affiliated
people writing half the commits.
Solid 1.9.14 defined fine-grained signals for the entire industry and runs on 91 commits a year,
one maintainer and about $29,000.
Comparison
| Version | Released | npm/week | Stars | Commits/12mo | Contributors | Top author | |
|---|---|---|---|---|---|---|---|
| React | 19.2.8 | 2026-07-21 | 163,138,787 | 247,055 | 926 | 85 | 23% |
| Preact | 10.29.8 | 2026-08-01 | 28,528,601 | 38,807 | 185 | 20 | 70% |
| jQuery | 4.0.0 | 2026-01-18 | 17,566,690 | 59,788 | 86 | 10 | - |
| Vue | 3.5.41 | 2026-08-05 | 14,393,209 | 54,128 | 435 | 100 | 31% |
| Lit | 3.3.3 | - | 6,958,082 | 21,754 | 70 | 19 | 47% |
| Angular | 22.1.0 | 2026-07-29 | 5,881,229 | 100,952 | 4,044 | 232 | - |
| Svelte | 5.56.8 | - | 5,366,553 | 87,910 | 796 | 109 | 30% |
| Solid | 1.9.14 | 2026-07-01 | 3,455,143 | 35,808 | 91 | 24 | 66% |
| Backbone | 1.6.1 | 2025-04-01 | 1,284,993 | 28,101 | 6 | 3 | - |
| Alpine.js | 3.15.12 | - | 660,921 | 31,840 | 132 | 29 | 53% |
| htmx | 2.0.10 | 2026-04-21 | 216,335 | 48,901 | 112 | 25 | 67% |
Angular’s npm figure is lower than Lit’s, which tells you more about transitive dependencies than about adoption. Preact’s 28.5M is double Vue’s and is almost entirely analytics scripts and embedded widgets.
React
19.2.8 (2026-07-21). 247,055 stars, 163.1M weekly downloads. One major release in three years -
19.0.0 landed 2024-12-05 - which is the slowest breaking cadence of the big five by a wide margin.
Governance changed, and the two dates are different. Intent was announced 2025-10-07, with Meta committing “over $3 million” across a five-year partnership plus full-time engineers. The React Foundation formally launched on 2026-02-24 at the Linux Foundation Member Summit, with eight platinum founding members - Amazon, Callstack, Expo, Huawei, Meta, Microsoft, Software Mansion, Vercel - and Seth Webster as executive director. Technical governance is explicitly independent of the board, though as of the launch only a provisional leadership council existed and no permanent charter has been published.
The concrete proof the transfer executed: github.com/facebook/react 301-redirects to
github.com/react/react, and the new org holds react, react-native, jsx, yoga, metro and
create-react-app. Meta’s namespace no longer owns React.
React Compiler 1.0 shipped 2025-10-07 - opt-in automatic memoization, not a change to the rendering model. React remains virtual-DOM based and is the only major framework that did not adopt signals, choosing a compiler that automates memoization instead.
RSC had a bad security year: two advisories in nine days, on 2025-12-03 (critical) and 2025-12-11, the second covering denial of service and source code exposure through the server rendering boundary. That is precisely the failure mode RSC critics predicted, now on the record from React’s own team.
Maintenance is healthy - 926 commits from 85 contributors, about five people carrying it - but the backlog is the worst here in absolute terms: 806 open issues excluding PRs, 60% older than two years, the oldest from 2013.
One structural wart: React does not ship its own types. They live in DefinitelyTyped, so react and
@types/react version independently and skew. It is alone among the big five in this.
Create React App was sunset 2025-02-14.
Vue
3.5.41 (2026-08-05). 54,128 stars, 14.4M weekly npm - but 888M jsDelivr hits a month, 3.3x React’s,
because Vue is genuinely used from a script tag.
The last minor was 3.5.0 on 2024-09-03 - 23 months of patch-only releases. Many blog posts claim “Vue
3.6 released in early 2026”; the registry settles it: latest is 3.5.41 and rc is 3.6.0-rc.2, reached
2026-07-22 after fourteen betas. Vapor mode is feature-complete, opt-in and at release candidate, not
stable. A stable date is unverified - claims of Q4 2026 come from third-party blogs, not vuejs.org.
Bus factor 1. edison1105 wrote 133 of 435 commits (31%); renovate[bot] contributed 132, so bot
commits nearly equal the top human’s. The next human has 11. Evan You does not appear in the top ten. The
backlog is 558 open issues with 72% older than two years - the worst stale ratio here.
Vue itself is not company-owned, and VoidZero never covered it - VoidZero’s remit was Vite, Vitest, Rolldown, Oxc and Vite+. VoidZero was acquired by Cloudflare on 2026-06-04, with Evan You and the whole team moving across, Vite staying MIT and vendor-agnostic, and $1M committed to a Vite ecosystem fund administered by the Vite core team. Vue remains sponsor-funded.
TypeScript works but needs vue-tsc/Volar as a second checker, and generic components need the
<script setup lang="ts" generic="..."> escape hatch.
Angular
22.1.0 (2026-07-29). 100,952 stars, 5.9M weekly downloads.
Two majors per year, every year, for three years - 20.0.0 (2025-05-28), 21.0.0 (2025-11-19), 22.0.0 (2026-06-03) - with formal LTS tags going back to v4. The most predictable cadence in the field.
Maintenance is the healthiest here by a distance: 4,044 commits from 232 distinct non-bot contributors.
And the top two human contributors, JeanMeche (416) and SkyZeroZx (360), are community members
outranking every Googler. There is no single point of failure. Google backs it with no external foundation;
team size is not publicly stated.
Zoneless change detection is stable and default. Per Angular’s own roadmap, the default strategy is now
OnPush and ChangeDetectionStrategy.Default was renamed to Eager. Signals went stable in 2025, Signal
Forms graduated to stable in 2026, and Vitest replaced Karma as the primary test runner in v21. zone.js is
retained for existing apps but omitted from new ones. In progress: the tsgo compiler port, Angular Aria
headless components.
Angular has completed the most radical architectural migration of any major framework - zone-patching
dirty-checking to signals-driven fine-grained reactivity - without a rewrite-your-app moment. It also has
the best TypeScript ergonomics by design intent, with one friction point: templates are checked by a
separate strictTemplates pass rather than by tsc directly.
Svelte
5.56.8. 87,910 stars - third-highest here, ahead of Vue’s core repo, which illustrates that stars measure
affection rather than deployment. 5.4M weekly downloads.
5.0.0 landed 2024-10-19 and there have been 56 minor releases in the 22 months since, with no major and no Svelte 6 announced anywhere. This is continuous delivery wearing semver clothing.
The current frontier is async: experimental await in components (5.36.0), experimental async SSR
(5.39.0), out-of-order rendering (5.43.0), createContext (5.40.0). SvelteKit 3 is in preview.
Two people write half of it: Rich-Harris 239 commits (30%) and dummdidumm 148 (19%). Vercel has
employed Rich Harris full-time since November 2021 and states governance stays with the independent
project. Practically, the two people producing half the commits are Vercel-affiliated, so Svelte’s velocity
is a Vercel budget line. The backlog is the best here - 41% older than two years, against Vue’s 72%.
TypeScript improved materially once runes gave the compiler real declaration sites, but .svelte files
still need svelte-check rather than plain tsc.
Solid
1.9.14 (2026-07-01). 35,808 stars, 3.5M weekly downloads - though Solid is the fastest-inflating package
measured, at 6.8x over fifteen months, so read that number with suspicion.
The last minor was 1.9.0 on 2024-09-24, 22 months ago. Solid 2.0 has been in prerelease since
2025-02-13 and is at 2.0.0-beta.31 after 49 prereleases, bringing first-class async, new control flow
and breaking changes to lists, effects, stores and directives.
The resourcing is the story. 91 commits in the last year. 24 contributors. ryansolid wrote 66% of them
and the next contributor has three. Last commit 2026-07-01. Its Open Collective shows an estimated annual
budget of $28,907, with Netlify, JetBrains, Builder.io and Sentry as top backers.
Solid has the best TypeScript ergonomics of any framework here - plain TSX checked directly by tsc, types
shipped in-repo, no separate checker, no @types/ split, no template language.
Solid won the architecture argument and lost the resourcing one. The fine-grained signal model it defined now ships inside three frameworks funded by Google, Cloudflare and Vercel.
Reactivity models
| Model | |
|---|---|
| React | Virtual DOM re-rendering whole subtrees on state change, with memoization now inserted automatically by React Compiler |
| Vue | Proxy-based fine-grained signals driving a compiler-optimised virtual DOM; Vapor mode compiles the VDOM away entirely |
| Angular | Signal-based fine-grained reactivity with OnPush change detection, zone.js out of the default path |
| Svelte | A compiler turning runes into fine-grained signal subscriptions that emit direct DOM mutations - no VDOM, no runtime diffing |
| Solid | Pure fine-grained signals; components run exactly once and reactive primitives update individual DOM nodes |
| Preact | Compact virtual DOM with optional signals that bypass the diff |
| Lit | Tagged template literals marking dynamic ranges once, updating only those ranges, over native Web Components |
| Qwik | Fine-grained signals plus resumability - serialising execution state into HTML so the client resumes rather than hydrates |
| Alpine.js | Reactive proxies bound to DOM attributes, evaluated in place with no build step |
| htmx | Not a reactivity model - attribute-declared AJAX swapping server-rendered HTML fragments into the DOM |
| Ember | Autotracking, a fine-grained invalidation system that predates and closely resembles signals |
Angular, Vue, Svelte 5, Solid, Preact, Qwik and Ember all converged on fine-grained signals. React is the sole holdout.
The TC39 Signals proposal is at Stage 1 and stalled - the README says Stage 1, it was presented for that stage in April 2024, and the repository has had three commits in the last twelve months, all on a single day in August 2025. Its own FAQ estimates native browser availability at “at least 2-3 years at an absolute minimum” after advancement. Every framework converged on signals; the standard did not follow. Do not plan around native signals.
Other frameworks
- Preact
10.29.8- 28.5M weekly downloads, almost all transitive through analytics scripts and widgets. Preact 11 has managed three betas in a year. - Lit
3.3.3- the web-components standard-bearer and the clearest slowdown here: 70 commits in 12 months, last commit 2026-06-23, with 716 open issues. Google-backed. High install base, low velocity. - Qwik
1.20.0- resumability instead of hydration. Qwik 2.0 moved to the@qwik.dev/corescope and sits at2.0.0-beta.38pulling 1,933 weekly downloads; combined with the legacy package that is ~44k, the smallest real footprint here. - Alpine.js
3.15.12- judge it by CDN, not npm: 625M jsDelivr hits a month, ranked 74th, 59 places above React. The default interactivity layer for Laravel/Livewire and the server-rendered world. - htmx
2.0.10- not a JS framework in the same sense. ~14KB of attributes that issue AJAX requests and swap server-returned HTML into the DOM: no component model, no client state, no build step. Your rendering framework is whatever renders on your server. Comparing it to React is a category error. htmx 4.0 is in beta and the project skipped 3.x entirely. - Stencil
4.43.5- a compiler emitting standards-compliant Web Components, with Stencil 5 in alpha. Mostly enterprise design systems. - Marko
6.3.32- technically impressive streaming SSR, and structurally the most fragile project here: 8 non-bot contributors, one writing 68%. - Mithril
2.3.8- dormant. 39 commits, last commit 2026-01-25, and a 3.0 that has been in prerelease since 2022 across 102 iterations.
Two projects widely written off that are not dead. Ember 7.1.0 shipped 7.0 and 7.1 in 2026 with
986 commits from 24 contributors - healthy engineering, terminal mindshare (State of JS 2024 had 62
respondents using it at work against React’s 8,548). And jQuery shipped 4.0.0 on 2026-01-17, its first
major in twelve years - dropping old IE, removing $.isArray/$.trim/$.type, migrating the source from
AMD to ES modules, and cutting over 3KB gzipped.
Backbone 1.6.1 is the actual corpse: 6 commits in a year from 3 people, yet still 1.28M weekly
downloads - more than Alpine, htmx, Ember, Mithril, Marko and Qwik combined. The purest illustration
available that npm downloads measure dependency graphs, not decisions.
What the surveys say
Stack Overflow 2025 (49,000+ responses, 177 countries) is the latest edition - there is no 2026 edition. React 44.7% overall and 46.9% among professionals, Angular 18.2%, Vue 17.6%, Svelte 7.2%, jQuery 23.4%.
The methodology point most write-ups miss: every single option rose year over year. When all options in a multi-select gain at once, the cause is the denominator - the sample fell from ~60k to ~26-32k on the technology section and skewed toward more engaged respondents picking more technologies each. The correct reading is ratios: React÷Vue went 2.56 → 2.54, React÷Angular 2.31 → 2.46, React÷Svelte 6.08 → 6.21. React’s relative position is flat. It is neither growing share nor losing it.
State of JS 2025 (~10,000-12,000 respondents) reports “remarkably stable over the past year, with no change in usage rankings”, except Alpine and htmx swapping places, and notes Solid holding the highest satisfaction five years running.
The counterweight is deployment data. The 2024 Web Almanac found jQuery on 74% of mobile and 75% of desktop pages crawled, against React’s 10%. There is no newer figure - the 2025 Web Almanac has no JavaScript chapter. Both things are true at once: React dominates new application development, jQuery dominates the web as it actually exists.
Both surveys are self-selected and skew heavily toward early adopters and English-speaking individual contributors, under-representing enterprise Angular and jQuery maintenance work.
Is there measurable migration away from React?
No. The criticisms are substantive - two RSC security advisories including source-code exposure, the involuntary Create React App migration, and a governance complaint serious enough that React changed its ownership structure over it. But no primary metric shows movement: the Stack Overflow ratios are flat, State of JS reports no ranking change, and the one framework growing faster on npm (Solid) is a 91-commit-a-year, single-maintainer, $29,000-a-year project with no stable 2.0 after 49 prereleases.
The measurable 2025-2026 changes in React’s world are institutional - the Foundation, the repository move, Compiler 1.0 - not demographic.