Benchmarks

Measured runtime and product workloads. Each table identifies its binaries, source, measurement method and execution host.

Source: CI run of the aphrody-perf workflow (artifacts perf-report-*).

linux-x64

Bun OS 1.4.6 (fd9d7a224) compared with upstream 1.4.3 (c6da4a4d3) on linux 6.17.0-1022-azure x64, AMD EPYC 7763 64-Core Processor x4, 16 GiB; 40 runs, 5 warmup, engine hyperfine, measured 2026-10-18T14:43:39.721Z. Median (p95); ratio = Bun OS / upstream, lower is faster for times. These measurements apply to the recorded executables and workload.

Recorded runtime measurements

Measure	Upstream	Bun OS	Ratio	Gate
bun --version	1.41 ms (p95 2.42 ms)	1.52 ms (p95 2.19 ms)	1.078	ok
bun -e ''	1.35 ms (p95 1.95 ms)	1.52 ms (p95 2.80 ms)	1.125	ok
bun run (empty script)	3.40 ms (p95 4.41 ms)	4.14 ms (p95 5.39 ms)	1.218	ok
bun test (one empty test)	4.10 ms (p95 5.25 ms)	5.33 ms (p95 6.21 ms)	1.301	ok
bun build (small project)	4.81 ms (p95 6.58 ms)	5.07 ms (p95 6.04 ms)	1.055	ok
bun install offline (warm cache)	4.02 ms (p95 4.68 ms)	4.36 ms (p95 6.10 ms)	1.085	ok
RSS at startup	13.3 MiB (p95 18.3 MiB)	29.1 MiB (p95 34.7 MiB)	2.185	info
RSS after loading node:* modules	27.4 MiB (p95 27.9 MiB)	43.9 MiB (p95 46.6 MiB)	1.606	info
require of node:* modules	37.5 ms (p95 42.8 ms)	35.5 ms (p95 36.9 ms)	0.947	ok
import() of node:* modules	37.5 ms (p95 43.4 ms)	36.1 ms (p95 37.2 ms)	0.963	ok
require of bun:* modules	1.22 ms (p95 1.64 ms)	1.52 ms (p95 1.83 ms)	1.253	ok
Bun.serve hello + first request	1.63 ms (p95 1.83 ms)	1.74 ms (p95 2.42 ms)	1.067	ok
local fetch latency (p50)	0.11 ms (p95 0.13 ms)	0.12 ms (p95 0.13 ms)	1.040	ok
local fetch latency (p99)	0.20 ms (p95 0.26 ms)	0.16 ms (p95 0.18 ms)	0.789	info
local fetch, sequential throughput	8,463 req/s (p95 9,065 req/s)	8,367 req/s (p95 9,902 req/s)	1.011	info
require of additional Bun OS modules	n/a	3.99 ms (p95 4.27 ms)		info
Binary size	84.9 MiB	184.9 MiB	2.179	ok

Run note: Release upstream bun-v1.4.6 absente : repli sur bun-v1.4.3. Versions de base différentes (fork 1.4.6, upstream 1.4.3) : l'écart inclut le travail amont entre les deux; taille et RSS ne bloquent pas (--strict pour les imposer).

What is measured

• bun --version
• bun -e ''
• bun run (empty script)
• bun test (one empty test)
• bun build (small project)
• bun install offline (warm cache)
• RSS at startup
• RSS after loading node:* modules
• require of node:* modules
• import() of node:* modules
• require of bun:* modules
• Bun.serve hello + first request
• local fetch latency (p50)
• local fetch latency (p99)
• local fetch, sequential throughput
• require of additional Bun OS modules
• Binary size

Thresholds live in bench/aphrody/thresholds.json: a case fails when the runtime median exceeds upstream by more than its ratio and its absolute delta.

Run it yourself

git clone https://github.com/aphrody-labs/bun && cd bun
# runtime binary against the upstream release of the same base version
bun scripts/aphrody/perf-gate.ts --fork ~/.bun/bin/bun --upstream-version 1.4.5 --runs 40 --out tmp/perf

The report is written to tmp/perf/perf-report.md and perf-report.json. The upstream micro-benchmarks of bench/ run with cd bench && bun install && bun run <name>.