Benchmarks

Every benchmark body we run, in one place. These are instruments, not trophy cases — a kernel Koru cannot express, a body without a live data feed, a number that got worse: all of it stays on this page, named.

7 benchmark bodies · 1 with live, reproducible boards. Bodies with a live board rank first; a write-up without one is a wiring gap, and it shows.

Compute kernels

live board · measured Oct 25, 2026, 05:53 PM

Faithful naive kernels — same algorithm, same parameters, no tricks — against C, Rust, Haskell and Osprey under one hyperfine protocol. Reference corpus and the Osprey column vendored from the Osprey language suite (MIT); measured on a shared, loaded M2 Pro, not a quiesced rig.

22 of 22 kernels ported 0 open gaps koru @a23463927 (dirty checkout)

Measured on a shared Apple M2 Pro under active load(load average 3.60 3.35 3.48on 12 cores) — not a quiesced benchmark rig. Read these as same-machine, same-protocol ballpark figures: run-to-run jitter on a loaded box routinely exceeds the gaps between the compiled languages, so treat close cells as ties, not rankings.

kernel	Koru	koru-fold	koru-par	C	cgcc3	cclang	cmap	Rust	Haskell	OCaml	Osprey	c3	odin	mojo	bend	bend-par	Koru / C
ackermann	21.8 ±0.2	—	—	128.2 ±1.7	—	127.8 ±1.0	—	131.6 ±1.6	75.5 ±2.5	—	128.2 ±1.6	—	—	146.3 ±1.7	58.0 ±1.3	—	0.17×
binarytrees	1.9 ±0.1	—	2.2 ±0.1	360.3 ±1.5	—	361.2 ±2.0	—	717.4 ±7.8	17.6 ±1.2	—	379.9 ±4.3	—	—	—	144.4 ±3.2	551.3 ±6.3	0.01×
coins	84.8 ±1.0	—	—	85.0 ±1.4	—	79.1 ±1.1	—	91.9 ±1.0	62.7 ±3.9	—	88.6 ±7.3	—	—	93.7 ±1.3	67.9 ±6.1	33.2 ±2.4	1.00×
collatz	14.7 ±0.3	—	3.6 ±0.2	12.3 ±0.8	—	12.3 ±0.2	—	14.6 ±0.4	41.1 ±1.7	—	15.1 ±0.2	—	—	26.8 ±1.5	53.0 ±5.3	48.4 ±2.8	1.20×
coprime	72.3 ±0.8	—	—	71.1 ±0.9	—	70.8 ±0.6	—	72.1 ±0.5	119.6 ±3.4	—	74.8 ±1.1	—	—	98.1 ±5.4	90.9 ±1.8	62.1 ±1.2	1.02×
digitsum	7.3 ±0.1	—	2.5 ±0.1	6.0 ±0.2	—	5.9 ±0.1	—	6.2 ±0.2	40.0 ±1.4	—	6.5 ±0.1	—	—	11.0 ±0.3	34.5 ±1.0	39.4 ±2.4	1.22×
exprtree	3.1 ±0.5	—	—	5.5 ±0.2	—	5.5 ±0.1	—	11.1 ±0.2	16.5 ±1.2	—	6.0 ±0.1	—	—	—	25.6 ±1.5	—	0.56×
factorial	40.0 ±0.3	—	6.7 ±0.7	40.1 ±0.6	—	40.2 ±0.6	—	41.4 ±3.7	63.5 ±2.1	—	40.4 ±0.6	—	—	56.6 ±0.4	609.5 ±7.6	491.8 ±15.7	1.00×
fib	27.3 ±0.3	—	38.8 ±15.0	27.4 ±0.3	—	27.1 ±3.1	—	27.6 ±1.0	63.3 ±1.7	—	27.5 ±0.3	27.7 ±3.1	27.7 ±0.6	31.3 ±0.3	68.5 ±5.3	29.2 ±2.1	1.00×
gcdsum	91.1 ±17.1	—	15.3 ±1.7	87.0 ±0.9	—	88.2 ±5.6	—	88.7 ±1.2	121.8 ±3.4	—	87.1 ±0.9	—	—	91.1 ±5.4	106.3 ±1.9	63.6 ±1.6	1.05×
hanoi	52.3 ±0.5	—	—	68.0 ±0.7	—	67.8 ±0.8	—	68.2 ±0.7	84.2 ±4.6	—	68.2 ±2.9	—	—	73.0 ±0.7	78.3 ±2.3	—	0.77×
isqrt	11.7 ±0.3	—	3.5 ±0.2	14.7 ±0.3	—	14.8 ±0.3	—	13.5 ±0.5	54.0 ±3.8	—	16.6 ±0.3	—	—	19.5 ±0.4	37.2 ±1.3	30.2 ±1.6	0.79×
josephus	36.5 ±4.5	—	—	35.6 ±0.7	—	36.0 ±0.5	—	35.5 ±0.3	52.8 ±1.8	—	36.3 ±0.7	—	—	53.5 ±1.1	55.5 ±4.3	—	1.03×
listops	—	—	—	1.4 ±0.1	—	1.4 ±0.1	—	1.8 ±0.2	17.0 ±2.5	—	4.5 ±0.1	—	—	—	19.2 ±1.4	—
mutual	16.6 ±2.4	—	3.9 ±0.3	16.9 ±0.8	—	15.8 ±0.7	—	17.2 ±0.8	40.7 ±2.3	—	16.1 ±0.5	—	—	6.4 ±0.3	21.8 ±1.9	21.7 ±1.8	0.98×
nestedloop	53.3 ±0.7	—	—	53.1 ±0.6	—	53.0 ±0.7	—	54.2 ±3.3	75.1 ±3.8	—	53.4 ±0.6	53.7 ±0.6	53.6 ±0.9	76.9 ±0.7	60.6 ±1.5	41.5 ±2.1	1.00×
pascal	33.8 ±0.4	—	—	33.7 ±0.4	—	34.0 ±0.5	—	34.0 ±0.5	87.2 ±2.1	—	34.0 ±0.6	—	—	39.4 ±0.6	255.4 ±3.5	68.2 ±6.9	1.00×
powmod	30.7 ±4.4	—	7.6 ±1.3	30.8 ±5.5	—	28.8 ±0.4	—	28.6 ±0.3	70.2 ±6.1	—	28.6 ±0.4	—	—	63.1 ±1.5	1515.8 ±12.5	816.5 ±1.2	1.00×
primes	7.9 ±0.2	—	2.8 ±0.1	8.0 ±0.2	—	7.9 ±0.1	—	8.6 ±0.2	28.5 ±1.1	—	8.0 ±0.2	8.6 ±1.1	8.2 ±0.2	12.9 ±0.3	28.0 ±1.0	25.2 ±1.6	0.99×
tak	36.3 ±0.3	—	36.7 ±0.5	43.0 ±0.6	—	42.9 ±0.3	—	43.3 ±0.5	101.9 ±4.4	—	43.2 ±0.6	—	—	47.5 ±0.3	103.2 ±6.0	50.6 ±3.8	0.84×
textstats	2.5 ±0.2	—	—	2.5 ±0.1	—	2.4 ±0.1	—	2.5 ±0.1	24.9 ±5.3	—	4.2 ±0.2	—	—	9.6 ±0.3	29.6 ±1.5	27.5 ±2.2	1.01×
wordfreq	4.0 ±0.1	1.6 ±0.1	—	1.7 ±0.1	—	1.7 ±0.1	2.8 ±0.1	5.8 ±0.3	16.6 ±0.8	—	20.8 ±0.8	5.7 ±0.7	4.7 ±0.3	11.3 ±0.6	21.1 ±1.8	25.4 ±3.4	2.36×

Mean wall-clock ms, lower is better · '—' = toolchain absent or no port · 'WRONG' = failed the oracle, excluded from timing · Koru / C is the ratio of means from this run only.

How these numbers were made

Machine: Darwin arm64 · Apple M2 Pro · load 3.60 3.35 3.48 on 12 cores (shared box, not quiesced)
Protocol: hyperfine -N, warmup 3, min 10 runs
Build flags: Koru: koruc build (ReleaseFast)
koru-fold: koruc build (ReleaseFast) — flat fold/identity port
koru-par: koruc build (ReleaseFast) — [parallel]-annotated port (fold split / fork-join)
C: cc -O2
cgcc3: gcc -O3 (real GCC only — ABSENT when gcc is clang)
cclang: clang -O3
Rust: rustc -C opt-level=3 -C overflow-checks=off
Haskell: ghc -O2
OCaml: ocamlopt -O3 -unsafe
Osprey: osprey --compile (release)
cmap: cc -O2 (hand-rolled string map)
c3: c3c compile -O5
odin: odin build -o:speed
mojo: mojo build (default -O3)
bend: bend build → native via clang (BendRT)
bend-par: bend build → native via clang — parallel-fork variant
Toolchains: C: Apple clang version 21.0.0 (clang-2100.0.123.102)
cgcc3: absent — not measured, never estimated
cclang: Apple clang version 21.0.0 (clang-2100.0.123.102)
Rust: rustc 1.96.0-nightly (b41f22de2 2026-03-08)
Haskell: ghc 9.10.3
OCaml: absent — not measured, never estimated
Osprey: osprey 0.0.0-dev
cmap: Apple clang version 21.0.0 (clang-2100.0.123.102)
c3: C3 Compiler Version: 0.8.3
odin: /Users/larsde/src/odin/odin version dev-2026-08:3e159a4
mojo: Mojo 1.2.0.dev2026091605 (9a2c5cbe)
bend: bend 2.0.5
bend-par: bend 2.0.5

MEASURED on the shared developer workstation named above, under live load (see machine.load_avg — the 1/5/15-min load average at run time) — NOT a quiesced benchmark rig. Read every number as a same-machine, same-protocol ballpark, not a precise cross-language ranking: run-to-run jitter on a loaded box routinely exceeds the gaps between the compiled languages. Wrong-answer binaries are excluded from timing. Absent toolchains are reported absent, never estimated. No cross-language claim leaves this board unless re-verified under the target's exact rules.

Osprey Kernel Suite: Koru Ports All 22 Kernels

Source, harness and raw results: github.com/korulang/koru-benchmarks