whatmodel.sh / bench

Benchmark your machine

[edit this page on GitHub ↗]

See how fast AI models run on your computer, and add your results to whatmodel.sh.

Run it

curl -fsSL https://whatmodel.sh/bench.sh | sh

Works on macOS and Linux. On Windows, run it in WSL. No Python? It offers to set it up for you. Want to read the script before it runs?

curl -fsSL https://whatmodel.sh/bench.sh -o bench.sh && less bench.sh && sh bench.sh

What happens

  1. It looks at your machine and the AI tools you already have.
  2. You tick what to test: models you have, models that fit your memory, and your subscriptions. Press / to search.
  3. It shows exactly what it will run, download and use, then waits for you.
  4. Tests run one at a time, in a random order. Ctrl-C stops at any time.
  5. Press Enter to upload your results, or upload them later with the command it prints.

Just looking? Try it with pretend models:

curl -fsSL https://whatmodel.sh/bench.sh | sh -s -- --demo

What's shared

Model names, speeds, runtime versions, and your chip, RAM and OS. Never file paths, prompts, answers, keys or your computer's name. Uploading opens a GitHub issue in your browser with your results filled in, and you press Submit yourself. If they're too long for a link, they're all put on your clipboard to paste in instead. They're released under CC0 and checked before they appear.

Launcher · Python source · Technical details · Other ways to contribute