Benchmarks

These are the headline results. Each table names its machine, its date and the result file in bench/results/. How they were measured is in Methodology.

Read this first

• The Benchmarks Game against Rust was re-run on 13 Nov 2026 (main 4a2f388), against Node and Bun the same day (main e5fb066). The other tables are from 5–10 Nov.
• Bun wins or ties some rows: the Todo API, energy per HTTP request, streaming time and SQLite time. Those rows are shown, not dropped.
• Against Rust no problem is lost on either VM. Two rows are ties within 5%: spectral-norm on both VMs, at the floor of the divisions the problem asks for. xisto's peak memory is at or below Rust's on every row (fannkuch-redux on the Xeon within 0.1 MB).

Benchmarks Game against Rust

Fastest matching variant on each side, median of 3, 13 Nov 2026, main 4a2f388, XISTO_CPU=native. Sources: bench/results/vm-2026-11-13/rust-ubuntu-xeon-2026-11-13-4a2f388.txt and bench/results/vm-2026-11-13/rust-arm-graviton2-2026-11-13-4a2f388.txt, scored by bench/placar.py.

machine	wins	ties (within 5%)	losses	memory
Xeon E5-2686 v4 (2 vCPUs)	9	1 (spectral-norm)	none	at or below Rust (fannkuch-redux 2.3 against 2.2 MB)
Graviton2 (2 vCPUs)	9	1 (spectral-norm)	none	below Rust on all 10

The per-problem tables are on Xisto vs Rust.

Benchmarks Game against Node and Bun

13 Nov 2026, the same xisto variants as the Rust round against the official Node programs under Node v26.12.0 and Bun 1.4.4, median of 3, on the Apple M4, the Xeon and the Graviton2. xisto wins every row on all three machines, on time and on memory: from 1.3× (regex-redux against Bun on the M4) to 200× (k-nucleotide against Node on the M4). Most xisto variants use every core while the Node programs use one, so the CPU columns are the fairer comparison; xisto uses less CPU on every row but regex-redux against Bun on the M4 (0.37 s against 0.34 s). The tables are on Xisto vs Node and Xisto vs Bun.

Servers

	xisto	Node	Bun	source
hello /ping, req/s	231,097	136,913	198,223	M4, 7 Nov, 2026-11-07-Apple-M4-http.csv
hello /ping, RSS after load	4.20 MB	100.72 MB	34.47 MB	same
Todo API list, req/s	111,914	68,697	115,425	M4, 5 Nov, 2026-11-05-Apple-M4-todo.csv
Conduit (NestJS) startup	21 ms	268 ms	—	M4, 8 Nov, conduit-mac-2026-11-08.txt
Conduit RSS at rest	2.6 MB	136.0 MB	—	same

What you ship

	xisto	Node	Bun
fib program	69,288 B	61,220,224 B	88,660,864 B

For Node and Bun the size is the runtime binary and its private libraries, which must be installed to run the program. Apple M4, 7 Nov 2026, bench/results/2026-11-07-Apple-M4.csv.

Pages

• Benchmarks Game: against Rust, Node and Bun, per problem.
• HTTP: hello, Todo API, the Conduit NestJS app, threads, resilience, slow dependencies.
• Energy, SQLite, streams and micro programs.

See also

• Methodology
• The performance model