Sounio Benchmarks

55 benchmark problems for the Sounio programming language — for evaluating LLM code generation quality.

HumanEval-Sounio (50 problems)

Classic coding interview problems translated to idiomatic Sounio:

Range	Topics
001-010	Two sum, FizzBuzz, string ops, Fibonacci, binary search, GCD, sieve
011-020	Sorting checks, merging, Roman numerals, balanced parens, spiral matrix
021-030	Determinant, median, moving average, histogram, Horner's method, numerical integration
031-040	Matrix multiply, Dutch flag, Boyer-Moore, stock profit, DP (stairs, robber, coins)
041-050	LIS, edit distance, unique paths, word break, set operations, inversions, permutations

Sounio-Specific Benchmarks (5 problems)

Problems that showcase Sounio's unique features:

File	Feature
sb_001	GUM uncertainty propagation
sb_002	Effect-tracked safe division
sb_003	Higher-order function pipeline (map/filter/fold)
sb_004	Running statistics accumulator (by-value struct update)
sb_005	Enum state machine with match

Running

export SOUC=./path/to/souc-linux-x86_64-jit
export SOUNIO_STDLIB_PATH=./path/to/stdlib
bash run_bench.sh

Results

All 55 problems pass souc run with exit code 0.

License

See Sounio main repository.