Benchmark comparing JavaScript bundlers and build tools (Rspack, Rsbuild, webpack, Vite, Rolldown, esbuild, Rollup, Parcel, Farm, Utoo and Bun) for dev server startup time, build performance and bundle size for applications with different module sizes.

• Build targets are Chrome >= 93 or es2022.
• Minification is enabled in production for all tools.
• Source map is enabled in development and disabled in production for all tools.
• Benchmarks run on GitHub Actions with variable hardware, which may cause inconsistent results.

Tooling details:

• webpack is configured to use SWC instead of Babel / Terser.
• Vite uses Rolldown and Oxc.

Data from GitHub Actions (2026-11-01, benchmark commit 8d085d4d).

A synthetic enterprise operations dashboard with 40 lazy routes and 29 common libraries, including React, Tailwind CSS, TanStack Router, Zod, and Recharts.

See case details.

CASE=react-dashboard pnpm benchmark

Development metrics:

Build metrics:

Dev memory (MiB):

Build memory (MiB):

A React app with 1,000 components and 1,500 modules from node_modules, using dynamic imports to simulate SPA.

CASE=react-1k pnpm benchmark

Development metrics:

Build metrics:

Dev memory (MiB):

Build memory (MiB):

A React app with 5,000 components and 5,000 modules from node_modules, using dynamic imports to simulate SPA.

CASE=react-5k pnpm benchmark

Development metrics:

Build metrics:

Dev memory (MiB):

Build memory (MiB):

A React app with 10,000 components and 10,000 modules from node_modules, using dynamic imports to simulate SPA.

CASE=react-10k pnpm benchmark

Development metrics:

Build metrics:

Dev memory (MiB):

Build memory (MiB):

A React app that imports UI components from several popular UI libraries.

Including @mui/material, @radix-ui/themes, antd, antd-mobile, @chakra-ui/react, @fluentui/react, @headlessui/react, @mantine/core, react-bootstrap, primereact, rsuite, @arco-design/web-react, @coreui/react, element-plus, ant-design-vue, naive-ui, primevue, vant, and vuetify.

CASE=ui-components pnpm benchmark

Build metrics:

Build memory (MiB):

A browser app that imports a small number of live exports from 50 popular, modern frontend libraries to compare tree-shaking quality across bundlers.

It keeps the original React/Vue/state/data set and adds 30 more mainstream frontend packages with ESM-friendly entry points where practical, including axios, dayjs, clsx, tailwind-merge, class-variance-authority, i18next, react-i18next, vue-i18n, immer, swr, framer-motion, three, lucide-react, @headlessui/react, @headlessui/vue, @heroicons/react, @heroicons/vue, @radix-ui/react-slot, query-string, mitt, fuse.js, idb, dexie, ky, lit, xstate, preact, solid-js, swiper, and remeda.

CASE=popular-libs pnpm benchmark

Build metrics:

Build memory (MiB):

Run the benchmark.ts script to get the results (requires Node.js >= 22):

# Run the benchmark for the react-5k case
pnpm benchmark

# Run the benchmark for the react-10k case
CASE=react-10k pnpm benchmark

If you want to start the project with the specified tool, try:

pnpm i # install dependencies

# Cd to the case directory
cd cases/react-5k
cd cases/react-10k
cd cases/popular-libs

# Dev server
pnpm start:rspack # Start Rspack
pnpm start:rsbuild # Start Rsbuild
pnpm start:webpack # Start webpack
pnpm start:vite # Start Vite
pnpm start:farm # Start Farm

# Build
pnpm build:rspack # Build Rspack
pnpm build:rsbuild # Build Rsbuild
pnpm build:webpack # Build webpack
pnpm build:vite # Build Vite
pnpm build:farm # Build Farm

Use CASE to switch the benchmark case:

CASE=react-1k pnpm benchmark
CASE=react-5k pnpm benchmark
CASE=react-10k pnpm benchmark
CASE=popular-libs pnpm benchmark

Use TOOLS to specify the build tools or bundlers:

# Run with all tools
TOOLS=all pnpm benchmark

# Run Rspack and Rsbuild
TOOLS=rspack,rsbuild pnpm benchmark

Use RUN_TIMES to specify the number of runs (defaults to 3):

RUN_TIMES=3 pnpm benchmark

Use WARMUP_TIMES to specify the number of warmup runs (defaults to 2):

WARMUP_TIMES=2 pnpm benchmark

Use FARM=true to run Farm:

FARM=true pnpm benchmark

Forked from farm-fe/performance-compare, thanks to the Farm team!

• Sample the command process tree every ~50 ms, excluding the browser and benchmark driver. Use macOS physical footprint (requires Xcode Command Line Tools) or Linux RSS; these metrics are not comparable across platforms.
• Measure timing and memory in separate passes, each with cold/warm caches and natural GC.
• Dev steady: load /, perform 10 root/leaf HMR updates, then idle for 5 seconds and take the median over 2 seconds. Peak: the largest sampled process-tree total during dev startup through this window, or during a build.
• Report MiB as the median across runs. Raw samples and summaries are saved in results/ and uploaded as CI artifacts.