scriptc benchmarks 1.6x faster than Bun --aot, author warns not to trust it
cramforce · x · 2026-10-18
scriptc is a new compiler that turns TypeScript/JavaScript into native executables or WebAssembly: it type-checks with tsc against your tsconfig.json, uses the type info to pick native representations, compiles via LLVM (npm dependencies included), and ships a single binary for macOS/Linux/Windows or WASM — no Node.js needed, with cross-compilation and coverage reports built in.
Malte Ubl (cramforce) benchmarked it against Bun's draft --aot mode using scriptc's own 30 workloads (5 interleaved runs, medians, whole-process timing):
- Overall: geometric mean 0.62x of Bun --aot (1.6x faster); 0.71x excluding 6 workloads using scriptc-native libraries that fall back to JS under Bun
- Resources: CPU 0.57x, peak memory 0.36x (mostly 11-47MB vs 22-96MB); binaries 121-527KB vs 128-133MB — Bun's are 250-1000x larger
- vs JIT: 0.71x of scriptc's own JIT and 0.41x of Node
- Where Bun --aot wins: cyclic-graphs 3x slower for scriptc (its tracing GC beats scriptc's reference counting + cycle collector), plus 3 more workloads
- scriptc's biggest wins: published-graphs 0.07x, intern-table 0.15x (first two use scriptc-native libraries)
The author adds a caveat: don't trust this benchmark — it's the one scriptc was optimized against and reflects neither your app nor any real one; the same applies to every benchmark you've seen.
More from coding & agent
- Leaked .env via agent leads to $90K Google bill — dotenv no longer safe for agentic dev — RealGeneKim · 2026-10-18
- supermemory ships npm plugins: one-command persistent memory for coding agents — gaganghotra_ · 2026-10-18
- Treating Cline Like an Engineering Team: PRD-Driven Development of an Open-Source Project — PSXP2W · 2026-10-18
- DeepSeek V4.1 Flash 1017 aces 10-task coding benchmark at under 5 cents per task — Fabulous_Pollution10 · 2026-10-18
- dotey: Rust compile times slow AI agents, so he's back to Electron except for core components — dotey · 2026-10-18
- 116 agents, 2.4B tokens: SI assistant 'passes the tests' by deleting them — tetsuoai · 2026-10-18