Benchmarks compare native Ax output with common baselines.

The benchmark suite is a regression harness for code generation and runtime overhead. It includes equivalent recursive, loop-heavy, string-search, path-normalization, URL-processing, CLI-argument, JSON-query, and simple HTTP backend programs for Ax, C, Rust, Python, and JavaScript.

Run

./benchmarks/run.sh

Every command is wrapped by benchmarks/expect-exit.sh, so Ax and each baseline must produce the same expected score before it is timed. The default run uses 5 warmups and 20 measured runs, and writes JSON plus Markdown reports to .ax-out/benchmarks/.

Recommended Tool

brew install hyperfine
./benchmarks/run.sh

Workloads

• fib: recursive integer function calls.
• mix: mutable while loop integer arithmetic with modulo.
• text: string length, prefix/suffix, and substring search loops.
• path: lexical path normalization and basename loops.
• url: URL parse, query lookup, and percent encode/decode loops.
• cli: command-line flag and option lookup loops.
• json: JSON compaction, validation, nested query, kind checks, and array lookup loops.
• web: simple HTTP backend route handling for /health and /ping.

Interpreting Results

Use benchmark results to catch regressions and compare implementation strategies. Do not treat one microbenchmark as a complete language ranking.