⚡ Performance

Benchmarked against Dyalog APL v20.0 and GNU APL v2.0

Benchmarked against Dyalog APL v20.0 and GNU APL v2.0 on Apple M5. Internal timing via ⎕AI (compute-only, no startup overhead). 40 iterations, 95% CI.

APL Run wins or ties 26 of 30 benchmarks vs Dyalog.

The AST viewer shows exactly which optimizations were applied to your code. Learn about execution badges →

Absolute Performance (msec, 95% CI)

Benchmark	Expression	APL Run	Dyalog	GNU APL
Prime Sieve
Primes N=2000	(2=+⌿0=∘.\\|⍨⍳2000)/⍳2000	1.8 ± 0.0	5.3 ± 0.9	65.6 ± 0.7
Primes N=4000	(2=+⌿0=∘.\\|⍨⍳4000)/⍳4000	7.4 ± 0.0	18.1 ± 1.0	—
Matrix Multiply
MatMul 100×100	M←100 100⍴⍳10000 ⋄ +/,M+.×M	0.2 ± 0.0	0.5 ± 0.2	5.6 ± 0.6
MatMul 200×200	M←200 200⍴⍳40000 ⋄ +/,M+.×M	0.9 ± 0.0	1.7 ± 0.3	39.1 ± 0.2
MatMul 400×400	M←400 400⍴⍳160000 ⋄ +/,M+.×M	7.4 ± 0.0	9.3 ± 0.5	—
Reductions & Arithmetic
Sum ⍳2M	+/⍳2000000` (★ fused to O(1))	0.0 ± 0.0	0.6 ± 0.6	10.0 ± 1.4
Sum ?2M⍴1000	+/?2000000⍴1000` (materialized)	8.7 ± 0.1	10.9 ± 0.5	32.5 ± 1.4
Arith chain 10M	+/(3×⍳10000000)+(-⍳10000000)	9.4 ± 0.1	11.2 ± 0.8	185.4 ± 3.2
Recursive Dfns
Collatz ⍳1K	+/c¨⍳1000` (recursive dfn)	1.3 ± 0.0	38.4 ± 1.9	—
Collatz ⍳5K	+/c¨⍳5000` (recursive dfn)	8.5 ± 0.4	216.7 ± 2.6	—
Fibonacci ⍳25	+/f¨⍳25` (doubly-recursive dfn)	7.1 ± 0.5	324.6 ± 4.1	—
Sort & Newton
Sort Int 100K	v[⍋v]` (100K random ints)	1.3 ± 0.1	1.9 ± 0.3	24.6 ± 0.3
Sort Float 1M	v[⍋v]` (1M random floats)	16.0 ± 0.2	17.7 ± 1.3	—
Sort String 100K	M[⍋M;]` (100K×10 char matrix)	15.3 ± 1.6	18.4 ± 1.3	—
Newton √100K	{⍵-((⍵*2)-N)÷2×⍵}⍣20⊢N	2.4 ± 0.2	4.7 ± 0.7	—
Newton √1M	{⍵-((⍵*2)-N)÷2×⍵}⍣20⊢N	22.7 ± 0.5	23.9 ± 1.1	—
LCS (Longest Common Substring)
LCS N=80	nested dfns, trains, ¨	3.4 ± 0.4	7.3 ± 1.2	—
FFT (Cooley-Tukey)
FFT 1024	recursive dfn, complex ×/+	2.1 ± 0.1	3.8 ± 0.3	—
FFT 4096	recursive dfn, complex ×/+	8.0 ± 0.2	13.8 ± 0.8	—
Standard Primitives
Transpose 3K×3K	⍉M` (3000×3000 matrix)	17.2 ± 1.3	8.6 ± 1.2	67.3 ± 3.5
Reverse 10M	⌽v` (10M-element vector)	5.8 ± 0.4	3.5 ± 0.9	66.0 ± 1.7
Membership 2M∊1M	a∊b` (2M in 1M)	3.9 ± 0.1	3.8 ± 0.5	—
Rotate 10M	100⌽v` (10M-element vector)	4.5 ± 0.4	3.2 ± 0.8	67.7 ± 2.0
Erdős Primes
Erdős ⍳2500	nested dfns, prime test, ¨	3.7 ± 0.0	6.8 ± 0.6	—
Erdős ⍳5000	nested dfns, prime test, ¨	8.0 ± 0.2	12.6 ± 0.7	—
Lucky Numbers
Lucky N=1600	recursive sieve, replicate, modulus	1.3 ± 0.4	0.8 ± 0.3	—
Lucky N=3000	recursive sieve, replicate, modulus	1.8 ± 0.0	2.1 ± 0.4	—
LSWRC (Unique Substrings)
LSWRC N=117	tacit defs, trains, compose, ¨	0.2 ± 0.0	0.4 ± 0.2	—
LSWRC N=567	tacit defs, trains, compose, ¨	2.5 ± 0.1	7.6 ± 0.8	—
Shannon Entropy
Shannon N=1500	Key (⌸), compose bind, tacit trains	3.6 ± 0.1	7.5 ± 0.5	—

Relative Performance (× slower than fastest)

Benchmark	APL Run	Dyalog	GNU APL
Primes N=2000	1.0×	2.9×	35.6×
Primes N=4000	1.0×	2.5×	—
MatMul 100×100	1.0×	2.4×	27.1×
MatMul 200×200	1.0×	1.8×	41.6×
MatMul 400×400	1.0×	1.3×	—
Sum ⍳2M	1.0×	6.0×	99.7×
Sum ?2M⍴1000	1.0×	1.3×	3.7×
Arith chain 10M	1.0×	1.2×	19.8×
Collatz ⍳1K	1.0×	29.2×	—
Collatz ⍳5K	1.0×	25.5×	—
Fibonacci ⍳25	1.0×	45.6×	—
Sort Int 100K	1.0×	1.5×	18.6×
Sort Float 1M	1.0×	1.1×	—
Sort String 100K	1.0×	1.2×	—
Newton √100K	1.0×	2.0×	—
Newton √1M	1.0×	1.1×	—
LCS N=80	1.0×	2.1×	—
FFT 1024	1.0×	1.8×	—
FFT 4096	1.0×	1.7×	—
Transpose 3K×3K	2.0×	1.0×	7.8×
Reverse 10M	1.7×	1.0×	18.9×
Membership 2M∊1M	1.0×	1.0×	—
Rotate 10M	1.4×	1.0×	20.8×
Erdős ⍳2500	1.0×	1.8×	—
Erdős ⍳5000	1.0×	1.6×	—
Lucky N=1600	1.7×	1.0×	—
Lucky N=3000	1.0×	1.2×	—
LSWRC N=117	1.0×	1.7×	—
LSWRC N=567	1.0×	3.1×	—
Shannon N=1500	1.0×	2.1×	—

Last benchmarked: 2026-07-18. APL Run is single-core (no threading). SIMD enabled. Recursive dfn benchmarks (Collatz, Fibonacci) use Dyalog mode. Sort uses ?N⍴N (integer data). Newton uses vectorized ⍣ iteration. LCS uses nested dfns with trains, Each (¨), and Unique (∪). Lucky Numbers uses a recursive sieve with replicate and modulus.