Benchmark evidence
Recorded fnpm comparisons with Bun, pnpm, npm and Yarn, including the cases fnpm loses.
fnpm’s target is to beat Bun in every package-management scenario while using less disk. Current evidence falls short of that target. Two recorded suites answer different questions and must stay separate.
Real repositories, five package managers
fnpm, Bun, pnpm, Yarn and npm each install the same pinned repositories, and each tree then runs the repository’s own check. Times are median seconds of three runs; disk is node_modules plus each tool’s private cache/store, hardlinks counted once. Every repository’s charts and numbers.
Measurement in progress: 2 repositories are complete for all five tools; OpenCode, Excalidraw, Jest, React, Vite, Vue core, webpack, Uptime Kuma are still running and appear here once every tool has finished.
Measured October 26, 2026 on Linux x64 (8 CPU cores, Node v24.22.1), fnpm built from 5e5b347. Each cell counts the repositories where fnpm's median was lower than the other tool's, out of the repositories where both installed.
| fnpm lower than | Bun | pnpm | Yarn | npm |
|---|---|---|---|---|
| Cold install | 1 / 2 | 2 / 2 | — | — |
| Warm cache | 1 / 2 | 2 / 2 | — | — |
| No-op | 2 / 2 | 2 / 2 | — | — |
| Disk | 2 / 2 | 1 / 2 | — | — |
Could not install: yarn on OpenNext Cloudflare, npm on OpenNext Cloudflare, yarn on tRPC, npm on tRPC. These pairs keep their rows in the charts, without a value.
Every completed install passed its repository's project check.
Choose a report
| Suite | Inputs | Statistic | What it tells you |
|---|---|---|---|
| Real projects | Pinned open-source repositories (OpenNext, OpenCode, tRPC, Excalidraw, Jest, React, Vite, Vue, webpack, Uptime Kuma) | Median of three successful measured runs | Install behavior of fnpm, Bun, pnpm, Yarn and npm on real projects, with package-set and direct project checks |
| Synthetic matrix | Small, medium, large, monorepo manifests | Mean of five measured runs plus one warmup | Controlled scenarios across fnpm, Bun, pnpm, npm and Yarn classic; includes the strict speed gate |
Both suites use Linux x64 and a pre-warmed local registry mirror. Timed installs disable lifecycle scripts. Process startup is included. OS and mirror caches remain warm even when a package manager’s private cache is empty. These are not public-registry or cross-platform performance claims.
What remains open
In the synthetic suite, the strict gate still fails 9/22 comparisons with Bun. Small cold installs, several cold-lockfile scenarios and all four add scenarios remain slower. Monorepo warm is also slower. fnpm’s combined disk footprint is lower than Bun and pnpm on all four synthetic fixtures; its extra cache/store alone is larger than Bun’s.
Read the methodology before interpreting either result. An install finishing or a project check passing covers only the command actually recorded. Full conformance and production readiness are not established.
What the timing excludes
The measurements do not time lock preparation, project builds, test suites or lifecycle scripts. Each suite records the fnpm source revision it measured. The synthetic suite measured an fnpm 0.0.1 source snapshot.