Benchmark Any Command in Seconds with Hyperfine

// Before you start

What you need

A terminal (Mac, Linux, or Windows)
Admin/sudo access for installation
At least one command you want to time

Step 1 of 6

Install hyperfine

2 min

Pick the one-liner that matches your operating system. Each command downloads and installs hyperfine from your system's official package repository. You only need to run one of these.

Terminal · mac

$ # Ubuntu or Debian

$ sudo apt install hyperfine

$ # Fedora

$ sudo dnf install hyperfine

$ # Arch Linux

$ sudo pacman -S hyperfine

$ # Alpine Linux

$ sudo apk add hyperfine

$ # macOS (Homebrew)

$ brew install hyperfine

$ # Windows (Winget)

$ winget install hyperfine

What you should see

Installation completes with no errors. You return to a normal prompt.

This might happen

apt install hyperfine says package not found on older Ubuntu

Download the .deb file directly instead: wget https://github.com/sharkdp/hyperfine/releases/download/v1.20.0/hyperfine_1.20.0_amd64.deb && sudo dpkg -i hyperfine_1.20.0_amd64.deb

Step 2 of 6

Confirm the install worked

1 min

Run a quick version check so you know hyperfine is on your PATH before you do anything else.

Terminal · mac

$ hyperfine --version

What you should see

hyperfine 1.20.0 (or similar version number)

This might happen

command not found after install

Close your terminal and open a new one, then try again. On Windows, restart the terminal entirely.

Step 3 of 6

Run your first benchmark

1 min

This times the built-in sleep command for 0.3 seconds. It is a safe, harmless test that shows you exactly what hyperfine output looks like. Hyperfine will run the command at least 10 times and for at least 3 seconds, then report statistics.

Terminal · mac

$ hyperfine 'sleep 0.3'

What you should see

A progress bar appears, then a summary table showing mean time (~300 ms), standard deviation, min, and max across all runs.

Step 4 of 6

Compare two commands head-to-head

2 min

Pass two commands as separate quoted arguments. Hyperfine runs both, then prints a comparison table showing which is faster and by how much. Replace the two example commands with any two commands you actually want to compare.

Terminal · mac

$ hyperfine 'find . -name "*.txt"' 'ls -R | grep ".txt"'

What you should see

Two rows in the results table, each with mean/min/max times and a 'Relative' column showing the speed ratio between them.

This might happen

One command errors out mid-benchmark

Test each command on its own in the terminal first to make sure it runs cleanly, then pass both to hyperfine.

Step 5 of 6

Add warmup runs for disk-heavy commands

2 min

Commands that read files are faster the second time because the OS caches the data. Adding --warmup 3 makes hyperfine run the command 3 times before it starts measuring, so your results reflect a warm cache rather than a cold one. This gives more realistic numbers for everyday use.

Terminal · mac

$ hyperfine --warmup 3 'grep -r TODO .'

What you should see

Hyperfine shows a 'Performing warmup runs' line before the benchmark starts, then the usual results table.

Step 6 of 6

Export results to a Markdown table

1 min

Add --export-markdown results.md to save a formatted table to a file. You can paste this table directly into GitHub READMEs, Notion, or any Markdown editor. The file is created in whatever folder your terminal is currently in.

Terminal · mac

$ hyperfine --warmup 3 --export-markdown results.md 'find . -name "*.txt"' 'ls -R | grep ".txt"'

What you should see

Benchmark runs as normal. Afterwards, a file called results.md appears in the current folder containing a Markdown table.

This might happen

results.md is empty or missing

Make sure you have write permission in the current folder. Try running the command from your home directory: cd ~ then repeat.

// Status

cooked. baked. worked.

A working hyperfine install that can time any shell command, compare two commands side by side, and export results to a Markdown file you can share.

// the honest bit

The honest part

Hyperfine measures wall-clock time, not CPU time, so background activity on your machine can add noise. Results are most reliable on an otherwise idle computer. It cannot profile why something is slow — only how long it takes. The parameterized benchmark feature (--parameter-scan) is powerful but requires you to understand the command you are varying; it is not covered here.