When we first published this guide, the standard advice online was to survey a dozen frameworks and pick your favourite. A new JavaScript framework seemed to appear every month, and articles like ours dutifully listed them all.

That advice has aged badly, and so has this article. Several of the frameworks we originally covered are now dead or dying, which is exactly why we rewrote this guide. The market has consolidated around the tools that actually ship production websites, not the ones with the most hype. After seven years of building headless websites, commerce storefronts and applications, our view is that the real shortlist in 2026 is small. For most projects, the decision comes down to four or five serious options, and the right one is usually fairly obvious once you know what kind of project you're building.

So this is a full rewrite. Rather than describing every framework that exists, we'll tell you which ones we actually recommend to clients, which ones we've stopped recommending, and how we make the call on real projects. We've written it for technical and non-technical readers alike, so we'll explain terms as we go.

A quick refresher: what a headless front-end is

In a headless build, your content and commerce data live in back-end platforms with APIs: a headless CMS such as Sanity, Payload, Storyblok or Contentful, or a commerce platform such as Shopify or Centra. The front-end is a separate application that pulls data from those APIs and renders the experience your visitors see.

That separation is the whole point. You can redesign the front-end without migrating the CMS. You can serve a website, a mobile app and an in-store screen from the same content. And you can decide, page by page, whether it's built in advance for maximum speed or built fresh on each request to deliver personalised content based on the visitor's identity or browsing context.

One thing has changed in how we talk about this. In 2024, it still made sense to start the conversation with React versus Vue versus Svelte, the underlying libraries developers use to build interfaces. In 2026, the vast majority of serious headless projects are built on a meta-framework rather than directly on a library. The underlying library comes along for the ride. You choose a meta-framework, the layer on top that handles the practical machinery of a website: which URL shows which page, how pages get built, how data is fetched, and how it all gets deployed. So that's how we've organised this guide.

What's changed since we last wrote this

A few shifts explain why the old version of this article needed retiring rather than a light edit.

The field has consolidated. The State of JavaScript 2025 survey found respondents had used an average of just 1.7 meta-frameworks. The long tail of options we used to list, including Frontity, Wuxt and VuePress, is gone or irrelevant. Gatsby, once the default choice for static sites, is in maintenance under Netlify's ownership, and its own community discussion boards openly ask whether it's dead. We wouldn't start a new project on it, and we actively help clients move off it.

Sending less JavaScript won. All the frameworks we consider worth using today build pages on the server and send the browser ready-made HTML, with JavaScript only where a page actually needs it. The technologies have different names (React Server Components, Astro's islands, Svelte's compiler) but they're all expressions of the same idea: send less code, get faster pages, especially on mobile. The old either-or debate about how pages get built has softened too, because modern frameworks let you mix approaches within a single page.

Vite became the common foundation. Vite is the build tool that turns a developer's code into a finished website, and almost every framework here now uses it. It recorded a remarkable 98 per cent satisfaction score in State of JavaScript 2025. The practical upside is that skills and tooling now transfer between frameworks far more easily than they used to.

Remix disappeared into React Router. Remix, which we covered as a rising option, merged its framework features into React Router 7 at the end of 2024. React Router 8 shipped in June 2026 as a deliberately boring, stable release. The Remix name now belongs to a new project heading in a very different direction. Interesting to watch, but not something we'd put a client site on today.

AI-assisted development now influences the choice. This one rarely gets said out loud in framework guides, but it matters on real projects. Development teams everywhere, including ours, use AI coding tools daily, and frameworks with lots of public code, stable conventions and good documentation get noticeably better AI assistance. It's not a reason to pick a framework on its own, but it quietly widens the productivity gap between mainstream options and niche ones.

The options that matter in 2026

Next.js: still the default, with honest caveats

Next.js remains the most widely used meta-framework by a distance, at roughly 59 per cent usage in State of JavaScript 2025, and it's still the framework we reach for most often on complex headless builds. Version 16, released in October 2025, tackled two long-standing complaints. It made Turbopack the default build tool, with the new filesystem caching feature available as an opt-in, which can compile sites two to five times faster. Earlier versions made confusing decisions on your behalf about which content to store and reuse; the new model, called Cache Components, keeps everything fresh unless you deliberately choose to cache it.

The caveats are real, though. Next.js attracted more complaints than any other framework in the latest State of JavaScript survey, mostly about complexity, and concerns about depending on Vercel have grown. Our experience is that Next.js rewards teams who invest in understanding it and punishes teams who treat it casually. Self-hosting is entirely doable, and we deploy Next.js to several platforms beyond Vercel, but it takes more care than the marketing suggests.

We recommend it for: complex sites and applications, enterprise headless builds with heavy integrations, and any project where hiring familiarity and ecosystem depth matter. It pairs well with Sanity, Payload, Storyblok and Contentful, all of which treat Next.js as a first-class citizen.

Astro: our pick for content-led websites

Astro has gone from interesting newcomer to our usual recommendation for content-driven websites, and the wider market seems to agree: it tops satisfaction ratings among meta-frameworks in recent developer surveys. It's trick is sending plain, fast HTML by default and only loading JavaScript for the parts of a page that actually do something, like a search box or a product carousel. Astro calls these islands, and the result is the kind of page speed scores in Google's Core Web Vitals that other frameworks have to fight for.

Astro 6, which reached stable release this year, made the version of a site developers run on their own machines behave much more like the live one, which catches problems earlier, and deepened its integration with Cloudflare's hosting platform. Live content collections now let content update without rebuilding the whole site, removing one of the classic annoyances of this style of framework.

It's worth noting that Cloudflare acquired Astro in January 2026, which explains the deepening integration. The same lock-in questions that apply to Next.js on Vercel or Nuxt after the NuxtLabs acquisition are worth asking here too, though Astro deploys to any platform and the project remains open source.

Two things worth knowing. First, Astro isn't the right tool for experiences that behave like apps, where lots of parts of the page need to talk to each other constantly; that's not a flaw, it's the design. Second, its ability to mix in React, Vue or Svelte components is genuinely useful for migrations but best used sparingly on brand-new builds.

We recommend it for: marketing websites, publishers, documentation and any headless CMS build where content is the product. If your site is mostly words and images with islands of interactivity, Astro is very hard to beat.

SvelteKit: the developer experience choice

Svelte's big idea is doing the heavy lifting when a site is built rather than in the visitor's browser, which means developers write less code and visitors download less of it. Svelte 5 rebuilt the framework's core around a new system it calls runes, and SvelteKit, the meta-framework built on top, remains the most pleasant one we work with. Performance is excellent without much effort.

The honest trade-off hasn't changed: the talent pool and ecosystem are smaller than React's. For our agency partners, this matters less than they expect, because we carry the SvelteKit expertise, but for an end client planning to take a build in-house, hiring is a fair consideration.

We recommend it for: teams who value speed of development and fast sites, interactive projects that don't need the React ecosystem, and clients who quite like the idea of something their competitors' sites aren't built on.

Nuxt: the obvious answer for Vue teams

Nuxt 4 is a mature, well-designed framework, and if your team already thinks in Vue, there's little reason to look elsewhere. The big news was corporate: NuxtLabs joined Vercel in July 2025, which raised the same lock-in questions the Next.js community wrestles with. So far the framework has stayed open and multi-platform; Nuxt 5 and the Nitro 3 server engine are in testing, and the development pace has, if anything, picked up.

We recommend it for: organisations with existing Vue investment. We wouldn't push a React-comfortable team towards it, but we'd never push a Vue team off it either.

Hydrogen: Shopify's toolkit for headless storefronts (evolving rapidly)

Shopify's relationship with Hydrogen just shifted. Until June 2026, it was an opinionated React framework bundled with the Remix router, designed as an all-in-one solution for headless Shopify builds. From June onwards, Shopify is rebuilding Hydrogen as a framework-agnostic toolkit, a set of commerce primitives (Storefront API client, cart logic, Shop Pay, analytics) that work with Next.js, Astro, SvelteKit, Nuxt or React Router, deployed to Oxygen, Vercel, Cloudflare Workers or standard Node.

Here's the honest assessment. The new Hydrogen is in developer preview, not production-ready. The API is subject to change, and customer accounts aren't yet supported. If you're building a new storefront today, we'd lean toward Next.js or Astro with Sanity for stable content tools that let you own your stack without volatility. If you're already on the original Hydrogen, it remains fully supported and production-grade.

We recommend it for: teams wanting Shopify's commerce primitives without framework lock-in, when the GA release ships. As it stands in 2026, we'd treat it as a preview, not a production choice. For Shopify Plus headless builds, the safer bet remains Next.js or Astro paired with Sanity or Payload, which keeps you on proven territory. The emerging Hydrogen toolkit will be worth revisiting once it reaches GA.

TanStack Start: the one to watch

The most credible new entrant in years, TanStack Start reached its 1.0 release candidate in late 2025 and drew hundreds of write-in mentions in State of JavaScript 2025 before it was even listed as an option. Its pitch: tooling that catches a whole class of mistakes before they reach visitors, no hidden magic, and no preferred hosting company. That lands well with developers who are tired of framework complexity and of being nudged towards one vendor's platform.

We're building with it internally and we like what we see so far. For client work we'd still class it as promising rather than proven, mainly because ecosystem maturity, CMS integrations and long-term maintenance patterns need time. Ask us again in a year.

What we no longer recommend

Worth being explicit here, because plenty of sites are still running on these. Gatsby is in maintenance and should be treated as end-of-life for new work; its best ideas live on in better form elsewhere. Frontity and Wuxt, the WordPress-specific frameworks we once covered, are discontinued. VuePress has been superseded by VitePress for documentation sites. And for headless WordPress generally, we now lean towards a standard framework like Next.js or Astro rather than specialised WordPress toolkits, because it keeps you on the well-trodden path.

If you're running a live site on any of these, no need to panic, but you should have a migration plan before ageing, unsupported code forces one on you.

How we actually choose on real projects

Framework guides usually end with "it depends", so here's what it actually depends on, in the order we weigh it.

Start with the content-to-application ratio. If the site is mostly content with pockets of interactivity, Astro. If it behaves like an application, with authentication, dashboards, personalisation or complex state, Next.js or SvelteKit. If it's a Shopify storefront with serious ambitions, Next.js or Astro with the Shopify Storefront API. Most projects resolve at this first question.

Then weigh the team. The best framework is one your developers, current and future, can work on confidently. React skills are the most abundant in the market, which is a legitimate reason Next.js wins ties. An existing Vue team points to Nuxt. An agency partner with no in-house developers can afford to be more adventurous, because their development partner carries the expertise.

Then check how well everything connects. Your CMS and commerce choices constrain the front-end less than they used to, but the quality of official integrations still varies, especially for things editors care about, like previewing a page before it goes live and editing content visually. Sanity, Payload, Storyblok and Contentful all support the mainstream frameworks well; the further you stray from the mainstream, the more custom code you'll write and maintain to wire things together.

Finally, think about longevity and lock-in. Ask who funds the framework, whether it runs well outside its sponsor's hosting platform, and what the last two major version upgrades were like. Boring, predictable releases, like React Router 8 or Nuxt's steady cadence, are a feature, not a disappointment.

Our take

The headless front-end market in 2026 rewards restraint. The interesting engineering has moved from choosing exotic frameworks to using mainstream ones well: making pages fast, keeping the code lean, and building publishing workflows your content team actually enjoys. In our experience, a well-built Astro or Next.js site with a well-structured headless CMS behind it beats a novel stack on speed, maintainability and cost far more often than not.

At Lemon Hive we've spent over seven years building headless websites, storefronts and applications on these frameworks, for end clients and as the development team behind other agencies. If you're weighing up a headless build, or you're sitting on a Gatsby site that needs a future, we're happy to talk it through. Get in touch.