smalljs$molRechercher⌘ KDocsPlaygroundComparerÉcosystèmeÀ proposFR

Svelte vs Vue

Svelte remporte 4 catégories sur 6, Vue 1, one à égalitéLes cas limites entrent dans le compte dès que vous lancez les tests en direct ci-dessous. Ils sont mesurés dans votre navigateur, pas stockés ici.

Cas limites

Pas encore lancé
Quatre situations ordinaires qui s’exécutent côte à côte dans votre navigateur. Appuyez sur « Lancer » sur un cas et lisez ce qui se passe ; rien ici n’est enregistré à l’avance.Une réponse périmée écrase la plus récenteCliquez dans une liste plus vite que le serveur ne répond.LancerSveltePas encore lancéVuePas encore lancéThe watcher assigns whatever the awaited request returns, with no cancellation. Ignoring a response that is no longer current takes a check after the await that the selected id has not changed.Afficher le code10 000 lignes de hauteur irrégulièreFaites défiler une longue liste où chaque ligne a une hauteur différente.LancerSveltePas encore lancéVuePas encore lancéEvery row is rendered by a plain v-for over the data. Rendering only the rows on screen takes a virtualization library and a way to measure each row, since the heights differ.Afficher le codeAbonnements après 100 cycles de montageMontez et démontez cent fois un composant abonné.LancerSveltePas encore lancéVuePas encore lancéonMounted subscribes and there is no matching onUnmounted. Releasing the subscription takes a teardown hook in every component that subscribes.Afficher le codeUn élément cassé fait tomber tout l’arbrePlacez des données mal formées dans une carte sur vingt.LancerSveltePas encore lancéVuePas encore lancéThe production build drops the throwing component and leaves an empty comment node in its place, so the rest of the list stays and the page shows no sign that anything failed — the TypeError reaches the console only. Putting something visible where the card was takes an onErrorCaptured hook and markup to render in the failed slot.Afficher le code

Code

Svelte 5 sur 5, Vue 0 sur 5 · Svelte devant
Combien il y a à lire, et quelle part appartient à quelqu’un d’autre.SvelteVueLines of code (TodoMVC)193247Svelte est 22% en dessous de VueFewer lines to read when someone else picks the project up. Counts the code you write and maintain yourself, not a stylesheet you installed from npm.Official TodoMVC implementation from github.com/tastejs/todomvc, plus the $mol one from the todomvc-compare repository. Every line of every source file under `src/`, tests, build config and the HTML shell excluded. Styles are excluded for every framework, because TodoMVC renders one fixed layout and its CSS is the same work regardless of framework: the tastejs examples install that stylesheet as the `todomvc-app-css` package and keep a small override sheet, while the $mol implementation writes its own. Counting styles would compare that packaging choice rather than the frameworks. The counting rule is the one published in todomvc-compare and reproduces its $mol figure to the line.Sveltegithub.com/…/svelte2026-08-13Vuegithub.com/…/vue2026-08-13npm packages shipped24Svelte est 50% en dessous de VueHow many npm packages end up inside the bundle the browser downloads. This is code your users execute, not code your build machine reads.Distinct `node_modules/<package>` paths in the `sources` of the production build sourcemaps. Scaffolded on 2026-08-13 with each framework's own getting-started command: react `npm create vite@latest app -- --template react-ts`; vue `npm create vue@latest app -- --default`; svelte `npx sv create app --template minimal --types ts --no-add-ons --install npm`; solid `npm create solid@latest app -- --vanilla --ts --template basic`; angular `npx @angular/cli@21 new app --defaults --skip-git --package-manager=npm`; $mol `git clone --depth 1 https://github.com/hyoo-ru/mam.git app && npm install`. For $mol, the sourcemap of the deployed todomvc.hyoo.ru bundle.Sveltesvelte.dev/…/creating-a-project2026-08-13Vuevuejs.org/guide/quick-start.html2026-08-13Dev dependencies declared710Svelte est 30% en dessous de VuePackages the starter project names in its own package.json. Build-time tooling included: this is the list your development environment inherits, not what reaches the browser.Entries in `dependencies` plus `devDependencies` of a fresh starter project. Scaffolded on 2026-08-13 with each framework's own getting-started command: react `npm create vite@latest app -- --template react-ts`; vue `npm create vue@latest app -- --default`; svelte `npx sv create app --template minimal --types ts --no-add-ons --install npm`; solid `npm create solid@latest app -- --vanilla --ts --template basic`; angular `npx @angular/cli@21 new app --defaults --skip-git --package-manager=npm`; $mol `git clone --depth 1 https://github.com/hyoo-ru/mam.git app && npm install`.Sveltesvelte.dev/…/creating-a-project2026-08-13Vuevuejs.org/guide/quick-start.html2026-08-13Dev packages installed22125Svelte est 82% en dessous de VueHow many separate packages npm resolves into node_modules on a clean checkout. Each one is an author your build trusts, and every CI run downloads all of them.Distinct name+version pairs in `npm ls --all` after scaffolding and installing the starter project. Scaffolded on 2026-08-13 with each framework's own getting-started command: react `npm create vite@latest app -- --template react-ts`; vue `npm create vue@latest app -- --default`; svelte `npx sv create app --template minimal --types ts --no-add-ons --install npm`; solid `npm create solid@latest app -- --vanilla --ts --template basic`; angular `npx @angular/cli@21 new app --defaults --skip-git --package-manager=npm`; $mol `git clone --depth 1 https://github.com/hyoo-ru/mam.git app && npm install`.Sveltesvelte.dev/…/creating-a-project2026-08-13Vuevuejs.org/guide/quick-start.html2026-08-13node_modules size68 MB98.3 MBSvelte est 31% en dessous de VueDisk footprint of a fresh checkout. None of it ships to the browser, but all of it is downloaded by every CI run and every new machine on the team.`du -sk node_modules` after scaffolding and installing the starter project. Scaffolded on 2026-08-13 with each framework's own getting-started command: react `npm create vite@latest app -- --template react-ts`; vue `npm create vue@latest app -- --default`; svelte `npx sv create app --template minimal --types ts --no-add-ons --install npm`; solid `npm create solid@latest app -- --vanilla --ts --template basic`; angular `npx @angular/cli@21 new app --defaults --skip-git --package-manager=npm`; $mol `git clone --depth 1 https://github.com/hyoo-ru/mam.git app && npm install`.Sveltesvelte.dev/…/creating-a-project2026-08-13Vuevuejs.org/guide/quick-start.html2026-08-13

Poids et chargement

Svelte 2 sur 2, Vue 0 sur 2 · Svelte devant
Ce que le navigateur télécharge, et combien de temps il attend avant que la page réagisse.SvelteVueTodoMVC bundle, gzip18.2 KB36.6 KBSvelte est 50% en dessous de VueWhat the browser actually downloads for a small but complete application, compressed the way a server would send it.Every JavaScript and CSS asset referenced by the production `index.html`, gzipped at level 9 and summed. Unlike the line count, styling is included everywhere and is comparable: the shared `todomvc-app-css` is compiled into each tastejs example’s shipped CSS, and $mol compiles its styles into the JavaScript bundle, leaving its `web.css` a 38-byte comment. The `base.js` page furniture from `todomvc-common` is excluded everywhere, including its Angular equivalent. Built artifacts come from the tastejs repository; the $mol bundle from the deployed todomvc.hyoo.ru.Sveltegithub.com/…/svelte2026-08-13Vuegithub.com/…/vue2026-08-13Startup transfer146 KB195 KBSvelte est 25% en dessous de VueBytes pulled over the network before the benchmark application is usable. Same application in every column, so the difference is the framework.Lighthouse total byte weight for the js-framework-benchmark application, reported as benchmark 34_startup-totalbytes. Transfer size, not gzip specifically.Sveltenin-jin.github.io/…/table.html2022-02-05Vuenin-jin.github.io/…/table.html2022-02-05

Vitesse

Svelte 7 sur 7, Vue 0 sur 7 · Svelte devant
Le temps que mettent les changements à atteindre l’écran, et la mémoire retenue pendant ce temps.SvelteVueCreate 1,000 rows173 ms201 msSvelte est 14% en dessous de VueTime to put a thousand fresh rows on the screen. This is the wait after hitting a filter or opening a list.js-framework-benchmark 01_run1k, arithmetic mean of 10 runs.Sveltenin-jin.github.io/…/table.html2022-02-05Vuenin-jin.github.io/…/table.html2022-02-05Partial update977 ms1,146 msSvelte est 15% en dessous de VueTouching every tenth row of a thousand, sixteen times. This is where a framework either finds the changed rows or re-checks everything.js-framework-benchmark 03_update10th1k_x16, arithmetic mean of 10 runs.Sveltenin-jin.github.io/…/table.html2022-02-05Vuenin-jin.github.io/…/table.html2022-02-05Swap two rows173 ms178 msSvelte est 2.9% en dessous de VueTwo rows trade places out of a thousand. The ideal answer is two DOM moves, and the gap here shows who gets close to it.js-framework-benchmark 05_swap1k, arithmetic mean of 10 runs.Sveltenin-jin.github.io/…/table.html2022-02-05Vuenin-jin.github.io/…/table.html2022-02-05Remove one row36.2 ms37.9 msSvelte est 4.5% en dessous de VueDeleting a single row out of a thousand. The smallest possible change, so it measures the fixed cost of reacting at all.js-framework-benchmark 06_remove-one-1k, arithmetic mean of 10 runs.Sveltenin-jin.github.io/…/table.html2022-02-05Vuenin-jin.github.io/…/table.html2022-02-05Memory after load1.22 MB1.37 MBSvelte est 11% en dessous de VueHeap the framework occupies with nothing on the page yet.js-framework-benchmark 21_ready-memory, single run.Sveltenin-jin.github.io/…/table.html2022-02-05Vuenin-jin.github.io/…/table.html2022-02-05Memory with 1,000 rows2.14 MB3.47 MBSvelte est 38% en dessous de VueHeap once a thousand rows are on screen. On a phone this is the difference between staying alive in the background and being killed.js-framework-benchmark 22_run-memory, single run.Sveltenin-jin.github.io/…/table.html2022-02-05Vuenin-jin.github.io/…/table.html2022-02-05Startup blocking time10.1 ms16.5 msSvelte est 39% en dessous de VueHow long the main thread is busy during startup, so taps and scrolls do nothing.js-framework-benchmark 32_startup-bt (Lighthouse total blocking time), arithmetic mean of 4 runs.Sveltenin-jin.github.io/…/table.html2022-02-05Vuenin-jin.github.io/…/table.html2022-02-05

Intégré

Svelte 2 sur 10, Vue 0 sur 10 · Svelte devant
Si le framework le fournit, ou s’il faut aller chercher une bibliothèque.SvelteVueRouterOuiOuiles deux l’ontWhether URLs and navigation come from the framework project itself, or you go pick a library and own that choice forever.Built in means the framework project ships it itself: its own package, or one the same team maintains and publishes as official. A library from anyone else counts as no, however popular or de-facto standard it has become, which is why routing is a yes for Angular and a no for React. The same cut is applied to all six frameworks, and every value links to the official page it was read from. Yes when that project publishes and documents a routing solution of its own. No when its own docs send you to a third-party project instead.Sveltesvelte.dev/…/routing2026-08-13Vuerouter.vuejs.org2026-08-13LocalizationNonNonni l’un ni l’autre ne l’aTranslating an application is rarely planned at the start and always arrives later.Built in means the framework project ships it itself: its own package, or one the same team maintains and publishes as official. A library from anyone else counts as no, however popular or de-facto standard it has become, which is why routing is a yes for Angular and a no for React. The same cut is applied to all six frameworks, and every value links to the official page it was read from. Yes when that project publishes and documents localization of its own. An official CLI add-on that wires up somebody else’s i18n library is a no, because the library is still somebody else’s.Sveltesvelte.dev/…/paraglide2026-08-14Vuevuejs.org/api2026-08-14ThemingNonNonni l’un ni l’autre ne l’aDark mode and brand colours without inventing your own CSS variable convention.Built in means the framework project ships it itself: its own package, or one the same team maintains and publishes as official. A library from anyone else counts as no, however popular or de-facto standard it has become, which is why routing is a yes for Angular and a no for React. The same cut is applied to all six frameworks, and every value links to the official page it was read from. Yes when that project publishes and documents a theming mechanism of its own. No when the word does not appear in its official API reference, which is the page linked, so the absence can be checked directly.Sveltesvelte.dev/…/overview2026-08-13Vuevuejs.org/api2026-08-13List virtualizationNonNonni l’un ni l’autre ne l’aLong lists stay smooth because off-screen rows are never rendered. Without it, someone eventually bolts a library onto a list that already shipped.Built in means the framework project ships it itself: its own package, or one the same team maintains and publishes as official. A library from anyone else counts as no, however popular or de-facto standard it has become, which is why routing is a yes for Angular and a no for React. The same cut is applied to all six frameworks, and every value links to the official page it was read from. Yes when that project publishes and documents lazy or virtual rendering of long lists. No when no such component appears in its API reference.Sveltesvelte.dev/…/overview2026-08-14Vuevuejs.org/api2026-08-14Offline supportOuiNonSvelte oui, Vue nonA service worker set up by the framework rather than copied from a blog post.Built in means the framework project ships it itself: its own package, or one the same team maintains and publishes as official. A library from anyone else counts as no, however popular or de-facto standard it has become, which is why routing is a yes for Angular and a no for React. The same cut is applied to all six frameworks, and every value links to the official page it was read from. Yes when that project ships the service worker integration: a module it builds and registers for you. No when the docs only show where to drop your own sw.js and how to register it with the browser API, because that is the browser’s feature rather than the framework’s.Sveltesvelte.dev/…/service-workers2026-08-13Vuevuejs.org/…/tooling.html2026-08-14Server renderingOuiOuiles deux l’ontRendering to HTML on the server, for search engines and for the first paint.Built in means the framework project ships it itself: its own package, or one the same team maintains and publishes as official. A library from anyone else counts as no, however popular or de-facto standard it has become, which is why routing is a yes for Angular and a no for React. The same cut is applied to all six frameworks, and every value links to the official page it was read from. Yes when that project publishes and documents a server rendering API of its own. No when its published module list contains no such module.Sveltesvelte.dev/…/page-options2026-08-13Vuevuejs.org/…/ssr.html2026-08-13Tests without a DOMNonNonni l’un ni l’autre ne l’aComponents testable as plain objects, with no browser and no jsdom to boot up first.Built in means the framework project ships it itself: its own package, or one the same team maintains and publishes as official. The same cut is applied to all six frameworks, and every value links to the official page it was read from. Yes when a component can be instantiated and asserted on with no document in play — the test never reaches for one. No when a DOM has to exist before a component can render at all, which is what every official testing guide here prescribes when it tells you to install jsdom, happy-dom or a browser. The question is whether a DOM is required, not whether the runtime happens to have one available: a runtime that provides a document nobody asks for does not make the test a DOM test.Sveltesvelte.dev/…/testing2026-08-13Vuevuejs.org/…/testing.html2026-08-13Dependency injectionOuiOuiles deux l’ontSwapping a real service for a fake one in a test without touching the component that uses it.Built in means the framework project ships it itself: its own package, or one the same team maintains and publishes as official. A library from anyone else counts as no, however popular or de-facto standard it has become, which is why routing is a yes for Angular and a no for React. The same cut is applied to all six frameworks, and every value links to the official page it was read from. Yes when that project documents a way to hand a component a value from outside it and replace that value later, in a test or in another host. Context APIs count: the bar is a documented mechanism, not a particular shape of container.Sveltesvelte.dev/…/context2026-08-14Vuevuejs.org/…/provide-inject.html2026-08-14FormsOuiNonSvelte oui, Vue nonValidation, dirty state and submission handled by the framework instead of a fourth form library per team.Built in means the framework project ships it itself: its own package, or one the same team maintains and publishes as official. A library from anyone else counts as no, however popular or de-facto standard it has become, which is why routing is a yes for Angular and a no for React. The same cut is applied to all six frameworks, and every value links to the official page it was read from. Yes when that project documents a form API that goes past binding one input to one variable, covering at least submission handling or validation state. Two-way binding on its own is a no.Sveltesvelte.dev/…/form-actions2026-08-14Vuevuejs.org/…/forms.html2026-08-14Typed templatesOuiOuiles deux l’ontA typo in a template is caught by the compiler, not by a user.Built in means the framework project ships it itself: its own package, or one the same team maintains and publishes as official. A library from anyone else counts as no, however popular or de-facto standard it has become, which is why routing is a yes for Angular and a no for React. The same cut is applied to all six frameworks, and every value links to the official page it was read from. Yes when the framework’s official TypeScript documentation covers type checking of its template or view layer.Sveltesvelte.dev/…/typescript2026-08-13Vuevuejs.org/…/overview.html2026-08-13

Écosystème et marché

Svelte 2 sur 4, Vue 2 sur 4 · à égalité
La facilité à recruter pour ça, à faire une recherche dessus et à trouver une pièce toute faite.SvelteVueGitHub stars87,97554,230Svelte est 62% au-dessus de VueA popularity signal, and nothing more. It says how many people bookmarked the repository, not how many shipped with it.`stargazers_count` from the GitHub API for the repository that holds the framework's current source.Sveltegithub.com/sveltejs/svelte2026-08-13Vuegithub.com/vuejs/core2026-08-13npm downloads5,273,631 per week14,550,122 per weekVue vaut 2.8× SvelteThe closest thing to a usage count. Inflated by CI runs, but inflated the same way for everyone.npm registry downloads for the last full week, for the package you install to use the framework.Svelteapi.npmjs.org/…/svelte2026-08-13Vueapi.npmjs.org/…/vue2026-08-13Stack Overflow questions6,281107,927Vue vaut 17.2× SvelteHow likely it is that your problem was already asked and answered by someone else.Question count of the framework's Stack Overflow tag, via the Stack Exchange API.Sveltestackoverflow.com/…/svelte2026-08-13Vuestackoverflow.com/…/vue.js2026-08-13npm maintainers32Svelte est 50% au-dessus de VueHow many people can publish a release. A low number is a bus factor.Length of the `maintainers` array in the npm registry metadata of the framework's primary package.Svelteregistry.npmjs.org/svelte2026-08-13Vueregistry.npmjs.org/vue2026-08-13

Coût de possession

Svelte 0 sur 3, Vue 2 sur 3 · Vue devant
Ce que l’arbre de dépendances et l’historique des versions demandent au fil des années.SvelteVueAdvisories in shipped packages00les deux annoncent la même valeurKnown vulnerabilities in the packages that reach the browser. These execute on your users’ machines, which makes them a different class of risk from anything in the build toolchain.Each shipped package, at its installed version, queried against the OSV database. Counts packages carrying at least one advisory.Sveltesvelte.dev/…/creating-a-project2026-08-13Vuevuejs.org/guide/quick-start.html2026-08-13Advisories in the dev tree30Vue n’en a aucun, Svelte en a 3Known vulnerabilities in the packages a fresh checkout installs. These run on developer and CI machines rather than in the browser, but somebody still has to triage every one.Total reported by `npm audit` on the starter project, straight after scaffolding it. Scaffolded on 2026-08-13 with each framework's own getting-started command: react `npm create vite@latest app -- --template react-ts`; vue `npm create vue@latest app -- --default`; svelte `npx sv create app --template minimal --types ts --no-add-ons --install npm`; solid `npm create solid@latest app -- --vanilla --ts --template basic`; angular `npx @angular/cli@21 new app --defaults --skip-git --package-manager=npm`; $mol `git clone --depth 1 https://github.com/hyoo-ru/mam.git app && npm install`.Sveltesvelte.dev/…/creating-a-project2026-08-13Vuevuejs.org/guide/quick-start.html2026-08-13Breaking releases in 3 years10Vue n’en a aucun, Svelte en a 1Major versions shipped since 2023. Every one of them is a migration someone on the team pays for.Count of major semver versions of the framework's primary npm package first published in the last three years. Left blank for packages that do not use semver majors.Svelteregistry.npmjs.org/svelte2026-08-13Vueregistry.npmjs.org/vue2026-08-13

Comment c’est mesuré

Chaque valeur ici vient d’une mesure ou d’une source extérieure, et porte avec elle le lien et la date du relevé. Une métrique ne compte pour une catégorie que si les deux frameworks la publient : sinon, celui dont le tableau est le plus complet l’emporterait sur l’étendue de ses données plutôt que sur ce que disent les chiffres. Là où personne n’a encore mesuré, la ligne le dit au lieu de deviner.Méthodologie et sources sur GitHub

Proposer un framework

Il en manque un ? Nommez-le et ouvrez une issue. Ajouter un framework, c’est un fichier de données avec ses mesures et, pour les tests en direct, une page runner : les deux vivent dans le dépôt, à côté de ceux qui y sont déjà.Ouvrir une issue
Type to search the documentation.