尚未运行四个日常场景，在你的浏览器里并排运行。在某个用例上按“运行”，看看会发生什么；这里没有任何预先录好的东西。过期响应覆盖了最新响应点击列表的速度快过服务器回应。运行React○尚未运行The effect calls setState with whatever the request returns, with no cancellation. Ignoring a response that is no longer current takes a cancel flag or an AbortController inside the effect.Svelte○尚未运行查看代码10,000 行高度不一的列表滚动一个每行高度都不同的长列表。运行React○尚未运行Every row is rendered by a plain .map() over the data. Rendering only the rows on screen takes a virtualization library and a way to measure each row, since the heights differ.Svelte○尚未运行查看代码100 次挂载循环之后的订阅把一个带订阅的组件挂载和卸载一百次。运行React○尚未运行The effect subscribes and returns nothing. Releasing the subscription takes a cleanup function returned from every effect that subscribes.Svelte○尚未运行查看代码一个损坏的条目拖垮整棵树在二十张卡片中的一张里放入格式错误的数据。运行React○尚未运行With no ErrorBoundary above it, a throw during render takes the whole tree down. Keeping the rest of the list on screen takes an ErrorBoundary component around every part that may throw.Svelte○尚未运行查看代码
代码
React 5 项中占 0 项，Svelte 5 项中占 5 项 · Svelte 领先要读的东西有多少，其中又有多少属于别人。ReactSvelteLines of code (TodoMVC)312193Svelte 比 React 低 38%Fewer 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.Reactgithub.com/…/react2026-08-13Sveltegithub.com/…/svelte2026-08-13npm packages shipped32Svelte 比 React 低 33%How 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.Reactreact.dev/learn/build-a-react-app-from-scratch2026-08-13Sveltesvelte.dev/…/creating-a-project2026-08-13Dev dependencies declared97Svelte 比 React 低 22%Packages 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`.Reactreact.dev/learn/build-a-react-app-from-scratch2026-08-13Sveltesvelte.dev/…/creating-a-project2026-08-13Dev packages installed3522Svelte 比 React 低 37%How 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`.Reactreact.dev/learn/build-a-react-app-from-scratch2026-08-13Sveltesvelte.dev/…/creating-a-project2026-08-13node_modules size80.9 MB68 MBSvelte 比 React 低 16%Disk 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`.Reactreact.dev/learn/build-a-react-app-from-scratch2026-08-13Sveltesvelte.dev/…/creating-a-project2026-08-13
重量与加载
React 2 项中占 0 项，Svelte 2 项中占 2 项 · Svelte 领先浏览器下载了什么，以及页面开始响应前要等多久。ReactSvelteTodoMVC bundle, gzip75.7 KB18.2 KBSvelte 比 React 低 76%What 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.Reactgithub.com/…/react2026-08-13Sveltegithub.com/…/svelte2026-08-13Startup transfer274 KB146 KBSvelte 比 React 低 47%Bytes 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.Reactnin-jin.github.io/…/table.html2022-02-05Sveltenin-jin.github.io/…/table.html2022-02-05
速度
React 7 项中占 0 项，Svelte 7 项中占 7 项 · Svelte 领先改动到达屏幕需要多久，以及这期间占用了多少内存。ReactSvelteCreate 1,000 rows224 ms173 msSvelte 比 React 低 23%Time 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.Reactnin-jin.github.io/…/table.html2022-02-05Sveltenin-jin.github.io/…/table.html2022-02-05Partial update1,263 ms977 msSvelte 比 React 低 23%Touching 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.Reactnin-jin.github.io/…/table.html2022-02-05Sveltenin-jin.github.io/…/table.html2022-02-05Swap two rows1,505 ms173 msSvelte 比 React 低 89%Two 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.Reactnin-jin.github.io/…/table.html2022-02-05Sveltenin-jin.github.io/…/table.html2022-02-05Remove one row37.2 ms36.2 msSvelte 比 React 低 2.7%Deleting 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.Reactnin-jin.github.io/…/table.html2022-02-05Sveltenin-jin.github.io/…/table.html2022-02-05Memory after load1.44 MB1.22 MBSvelte 比 React 低 15%Heap the framework occupies with nothing on the page yet.js-framework-benchmark 21_ready-memory, single run.Reactnin-jin.github.io/…/table.html2022-02-05Sveltenin-jin.github.io/…/table.html2022-02-05Memory with 1,000 rows4.24 MB2.14 MBSvelte 比 React 低 50%Heap 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.Reactnin-jin.github.io/…/table.html2022-02-05Sveltenin-jin.github.io/…/table.html2022-02-05Startup blocking time102 ms10.1 msSvelte 比 React 低 90%How 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.Reactnin-jin.github.io/…/table.html2022-02-05Sveltenin-jin.github.io/…/table.html2022-02-05
内置
React 10 项中占 0 项，Svelte 10 项中占 2 项 · Svelte 领先这件事框架自带，还是需要去找一个库。ReactSvelteRouter没有有Svelte 有，React 没有Whether 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.Reactreact.dev/learn/creating-a-react-app2026-08-13Sveltesvelte.dev/…/routing2026-08-13Localization没有没有两边都没有Translating 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.Reactreact.dev/reference/react2026-08-14Sveltesvelte.dev/…/paraglide2026-08-14Theming没有没有两边都没有Dark 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.Reactreact.dev/reference/react2026-08-13Sveltesvelte.dev/…/overview2026-08-13List virtualization没有没有两边都没有Long 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.Reactreact.dev/reference/react2026-08-14Sveltesvelte.dev/…/overview2026-08-14Offline support没有有Svelte 有，React 没有A 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.Reactreact.dev/reference/react2026-08-14Sveltesvelte.dev/…/service-workers2026-08-13Server rendering有有两边都有Rendering 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.Reactreact.dev/…/server2026-08-13Sveltesvelte.dev/…/page-options2026-08-13Tests without a DOM没有没有两边都没有Components 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.Reactreact.dev/warnings/react-test-renderer2026-08-13Sveltesvelte.dev/…/testing2026-08-13Dependency injection有有两边都有Swapping 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.Reactreact.dev/learn/passing-data-deeply-with-context2026-08-14Sveltesvelte.dev/…/context2026-08-14Forms有有两边都有Validation, 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.Reactreact.dev/…/form2026-08-14Sveltesvelte.dev/…/form-actions2026-08-14Typed templates有有两边都有A 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.Reactreact.dev/learn/typescript2026-08-13Sveltesvelte.dev/…/typescript2026-08-13
生态与市场
React 4 项中占 3 项，Svelte 4 项中占 1 项 · React 领先招人、搜答案、找现成零件各有多容易。ReactSvelteGitHub stars247,21887,975React 是 Svelte 的 2.8×A 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.Reactgithub.com/react/react2026-08-13Sveltegithub.com/sveltejs/svelte2026-08-13npm downloads163,083,190 per week5,273,631 per weekReact 是 Svelte 的 30.9×The 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.Reactapi.npmjs.org/…/react2026-08-13Svelteapi.npmjs.org/…/svelte2026-08-13Stack Overflow questions473,9256,281React 是 Svelte 的 75.5×How 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.Reactstackoverflow.com/…/reactjs2026-08-13Sveltestackoverflow.com/…/svelte2026-08-13npm maintainers23Svelte 比 React 高 50%How 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.Reactregistry.npmjs.org/react2026-08-13Svelteregistry.npmjs.org/svelte2026-08-13
持有成本
React 3 项中占 1 项，Svelte 3 项中占 0 项 · React 领先依赖树和发版历史在这些年里要付出什么。ReactSvelteAdvisories in shipped packages00两边报告的数值相同Known 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.Reactreact.dev/learn/build-a-react-app-from-scratch2026-08-13Sveltesvelte.dev/…/creating-a-project2026-08-13Advisories in the dev tree03React 没有，Svelte 有 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`.Reactreact.dev/learn/build-a-react-app-from-scratch2026-08-13Sveltesvelte.dev/…/creating-a-project2026-08-13Breaking releases in 3 years11两边报告的数值相同Major 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.Reactregistry.npmjs.org/react2026-08-13Svelteregistry.npmjs.org/svelte2026-08-13
