JavaScript Meta-Frameworks

A meta-framework is, in State of JS’s definition, a framework “focused on rendering and serving your application” - routing, SSR, SSG and deployment on top of a UI library. The term is the ecosystem’s, not the projects’: none of them calls itself one. Next.js says “a React framework”, Nuxt says “full-stack framework”, SvelteKit just “framework”.

Figures come from the npm registry and GitHub on 2026-08-06. npm counts roughly doubled registry-wide between January and April 2026, so compare packages against each other in the same week, never against their own past.

Summary

Next.js 16.3.0 is the only one shipping React Server Components as a stable default, has annual majors on a published schedule, and is the deepest Vercel commitment on this page.

Astro 7.1.6 is content-first with islands, has the most contributors of any project here (228), and has been backed by Cloudflare since January 2026. No RSC, deliberately.

Nuxt 4.5.2 and SvelteKit 2.70.2 are the Vue and Svelte answers; both are Vercel-backed, and neither implements RSC - SvelteKit has ruled it out on the record.

React Router 8.3.0 absorbed Remix in 2024 and is now three products under two names. Remix 3 is a beta that contains no React at all.

TanStack Start 1.168.37 reached real distribution, sponsor-funded and independent.

Comparison

	Version	Released	npm/week	Stars	Contributors	Backing	RSC
Next.js	16.3.0	2026-08-03	54,035,511	141,635	196	Vercel	stable, default
React Router	8.3.0	2026-07-22	51,516,259	56,544	103	Shopify	unstable_ only
TanStack Start	1.168.37	2026-08-05	16,780,323	14,898	215	sponsors	experimental, off
Astro	7.1.6	2026-07-29	4,327,531	61,576	228	Cloudflare	no
SvelteKit	2.70.2	2026-07-29	2,450,323	20,719	137	Vercel	no, on the record
Nuxt	4.5.2	2026-08-05	1,954,409	60,726	170	Vercel	no
Docusaurus	3.10.2	2026-07-10	1,452,338	65,838	99	Meta	no
Gatsby	5.16.1	2026-02-10	294,343	55,936	13	Netlify	no

Next.js

16.3.0 (2026-08-03). 54.0M weekly downloads, 141,635 stars, 196 contributors.

The release policy is now documented. Majors ship annually, targeting around 26 October: 14.x (2023-10-26), 15.0 (2024-10-21), 16.0 (2025-10-21). The current major is Active LTS; Maintenance LTS runs two years from a major’s release, so 15.x support ends 2026-10-21 and 14.x is already unsupported.

Next.js 16 made Turbopack stable and default for both next dev and next build (opt out with --webpack), introduced Cache Components folding use cache, PPR and dynamicIO into one flag, renamed middleware.ts to proxy.ts on the Node runtime while deprecating Edge middleware, and removed next lint along with synchronous params/cookies()/headers(). At the announcement Vercel reported that “>50% of development sessions and 20% of production builds on Next.js 15.3+” were already on Turbopack - the one real telemetry number available.

16.3 cut Turbopack dev memory by 90%, turned on a filesystem build cache by default, added TypeScript 7 support for next build type checking, and switched App Router SSR from Web Streams to native Node streams for a claimed +22% under load.

The Pages Router is not deprecated, but it is frozen. It still receives security patches. Its docs carry lastUpdated values from 2024-11 to 2025-05 against 2026-06/07 for the App Router, and every new capability - Cache Components, use cache, PPR, Instant Navigations - is App-Router-only. The often-quoted Vercel commitment to support it “for many years” is from a 2023 discussion and was not restated in 2026. No first-party adoption telemetry exists for App versus Pages; percentages in circulation are unsourced.

Next ships and uses a bundled React Canary rather than the published stable, which is how App Router features land ahead of React releases.

Astro

7.1.6 (2026-07-29). 4.3M weekly downloads, 61,576 stars, and 228 contributors - the broadest base of any project on this page. Node 22.12+.

Backed by Cloudflare since 2026-01-16. Note the pattern: Cloudflare now backs Astro and, through the VoidZero acquisition, Vite - while Vercel backs Next, Nuxt and Svelte.

Two majors landed unusually close together: 6.0 on 2026-03-10 and 7.0 on 2026-06-22, only 3.5 months apart, because 7 is largely the stabilisation of what 6.x shipped experimentally.

• Astro 6 rebuilt the dev server on Vite’s Environment API so the production runtime runs in dev, which kills the “works in dev, breaks on Workers” class of bug. Plus a built-in Fonts API, stable Live Content Collections and stable CSP.
• Astro 7 replaced the Go .astro compiler with a Rust one, moved to Vite 8 + Rolldown, and introduced Sätteri, a Rust Markdown/MDX pipeline that is now the default. Claimed 15-61% faster builds, on Astro’s own benchmarks. It also added astro dev --background with automatic coding-agent detection and astro dev --json structured logs - no other major framework ships agent detection in its dev server.

Islands are the architecture: all client JavaScript is stripped unless a client:* directive applies, and server islands (stable since Astro 5) let dynamic fragments stream into a cached shell. It is UI-library-agnostic - React, Vue, Svelte, Solid and Preact all mount as islands - which is the main structural difference from everything else here.

Astro DB was removed in 7.0, with the team stating they no longer had the bandwidth and were “not as interested as we once were”. Astro Studio was shut down in 2025. Use Drizzle, Kysely, Prisma or node:sqlite.

Nuxt and SvelteKit

Nuxt 4.5.2 (2026-08-05). 1.95M weekly downloads, 60,726 stars, 170 contributors. NuxtLabs was acquired by Vercel on 2025-07-08. Supports SSG, SSR, ISR and per-route hybrid rendering; Vue server components remain experimental. Built on Nitro, which is itself mid-migration (see JavaScript Web Frameworks).

SvelteKit 2.70.2 (2026-07-29), with 3.0.0-next.14 in preview. 2.45M weekly downloads, 20,719 stars, 137 contributors. Vercel employs the core maintainers. Supports SSG, SSR, SPA and streaming, with remote functions experimental. It has ruled out React Server Components on the record - which for a Svelte framework is less a rejection than a statement that the RSC model is React-specific.

React Router, and the Remix confusion

There are three products under two names, and most write-ups get this wrong.

• 2024-05-15 - the merge is announced: “What we planned to release as Remix v3 is now going to be released as React Router v7.” Remix packages take “a little nap.”
• 2024-11-22 - React Router v7.0.0 ships.
• 2025-05-28 - “Wake up, Remix!” The Remix name is revived for a new framework that is not React at all - its own runtime, Fetch-API routes, no RSC.
• 2026-06-17 - React Router v8.0.0, a cleanup major: drops the react-router-dom package, ESM-only, ES2022, Node 22.22+, React 19.2.7+, Vite 7+.

End of life: v8 declared React Router v6 and Remix v2 EOL as of 2026-06-17, so neither receives security updates. React Router v7 is still supported - 7.18.2 shipped 2026-07-28, after v8.3.0. The @remix-run/* packages are frozen at 2.17.5 and still depend internally on react-router@6.30.4; they carry no npm deprecated flag, so tooling will not warn you.

Remix 3 is at 3.0.0-beta.5 with 24,519 weekly downloads and its own blog stating “not production ready yet.”

React Router moved to open governance in June 2025, with a steering committee and a five-stage RFC process

• the only project on this page to do so.

Other meta-frameworks

• TanStack Start 1.168.37 - 16.8M weekly downloads and 215 contributors, independent and sponsor-funded. SSR, SSG, SPA, streaming and server functions; RSC is experimental and off by default. A @tanstack/solid-start exists too.
• SolidStart 2.0.0 (2026-08-04) - 97,264 weekly downloads, 19 contributors. The old solid-start package from 2024 is dead; @solidjs/start is the live one.
• Analog 2.6.4 - Vite plus Nitro for Angular. Community-run; any official Angular-team endorsement is unverified.
• Qwik Router - @qwik.dev/router has been at 2.0.0-beta.38 and Qwik 2.0 has been in beta for roughly 2.5 years. Builder.io is a sponsor, not the owner, since March 2024.
• Docusaurus 3.10.2 - Meta-backed, 99 contributors, the default for documentation sites.
• VitePress 1.6.4 - no release in 12 months, with 2.0.0-alpha.19 in preview.
• Eleventy 3.1.6 - SSG only, no UI framework, and four contributors. Zach Leatherman joined Font Awesome in 2024.
• Waku 1.0.0-beta.8 - RSC-first with 'use client' as the opt-out. Never reached 1.0.
• RedwoodSDK 1.6.0 - RSC by default, but locked to Cloudflare Workers.

Wound down: Redwood GraphQL - @redwoodjs/core@8.9.0 (2025-10-21) announced the wind-down, and the project managed 33 commits from 8 contributors in the last year. The community fork CedarJS has 142 stars, which is a high-risk successor. Gatsby is not archived but managed 127 commits from 13 contributors, most of them dependabot in /starters. Aleph.js was archived 2025-07-06. Create React App was sunset 2025-02-14.

React Server Components outside Next.js

Six implementations exist. Only Next.js ships RSC stable and on by default. The two others with real distribution both ship it off by default and marked unstable.

	RSC	Stability	Default
Next.js	yes	stable	yes (App Router)
React Router v7/v8	yes	unstable_	opt-in
TanStack Start	yes	experimental	opt-in
Waku	yes	beta	yes
RedwoodSDK	yes	stable 1.x	yes, Cloudflare-only
Parcel	yes	beta, unchanged 17 months	opt-in
Expo Router	yes	experimental, 21 months	opt-in, two flags
Remix 3	no React at all	-	-
Astro, Nuxt, SvelteKit, Qwik, SolidStart	no	-	-

Vite did not ship RSC in core. It lives in @vitejs/plugin-rsc - official by organisation, at 0.5.33 after 60 versions without reaching 1.0. The enabler is Vite’s Environment API, which formalises separate client/ssr/rsc module graphs, and Vite’s own documentation still says “We don’t recommend switching to Environment API yet.”

The detail that gets least attention: Vite’s official RSC plugin runs on React’s webpack bindings. It peer-depends on react-server-dom-webpack; the roadmap items to split out react-server-dom-vite and stop using the webpack package are both still unchecked, and react-server-dom-vite has zero published versions on npm.

RSC also had a bad security year - two advisories in nine days in December 2025, one of them covering source code exposure through the server rendering boundary.