hyperfine 2.0 Released: Benchmarking Tool Adds Memory and CPU Cycle Metrics
charliermarsh · x · 2026-10-09
The command-line benchmarking tool hyperfine has released v2.0.0, maintained by sharkdp and shared by Rust developer charliermarsh.
Key features
- Support for alternative performance metrics: peak memory usage, instructions, CPU cycles, cache misses, branch misses, etc. By default it now shows wall-clock time and peak memory; use --metrics to select others.
- New --env flag to set environment variables without an intermediate shell, enabling parameterized benchmarks (e.g. comparing thread counts).
- Plotting and analysis scripts now support multiple metrics.
Breaking changes
- Benchmarked commands now run directly by default without an intermediate shell (--shell=none). Use -S to restore the old behavior (sh on Unix, cmd.exe on Windows) or --shell <SHELL>.
- --setup, --prepare, --conclude, and --cleanup commands always run in a shell.
More from coding & agent
- LlamaIndex Explains Why Markdown Is Its Default Output for AI Document Parsing — llama_index · 2026-10-09
- SWEeper-Bench measures whether AI agents can squash bugs before users ever see them; Grok 4.6 leads — liuzhuang1234 · 2026-10-09
- Jev-Style Classifiers: Letting Agents Build Fast, Cheap Classifiers Instead of Fine-Tuning — vesko_st · 2026-10-09
- Adding spatial memory boosts Opus 5.5's Minecraft agent progress by 133% — ZhitingHu · 2026-10-09
- Voice-controlling multiple coding agents: handoff is trickier than it looks — DanWahlin · 2026-10-09
- Ramp and Stripe build internal vibe-coding apps so non-engineers ship tools — The Pragmatic Engineer · 2026-10-09