What does 30 tokens a second feel like beside 100? llmspeed lets you watch the difference and race two hardware/model combinations in a browser.

The pictured comparison replays a fixed answer at estimated rates; it doesn't run either model on those machines. The October 17 update expanded the model and hardware catalogue and refreshed benchmark data, with Claude credited in the commits.

Check the Measured / Derived / Projected labels. Users appreciated the comparison but also reported estimates that missed their own results.

Screenshot of llmspeed; try it:
llmspeed.dev/