What one Tiiny actually does under load | A deeper dive into measured numbers, how they were taken, and how to get your own

A spec sheet does not tell you what a box does when you give it real work, so this is a set of numbers measured on one instead.

What is on the page: five models, 23 runs, measured between 15 August and 15 September. Every result carries the build, the host and the NPU cost it was measured with, because a figure taken off a busy device is worthless and there is no way to tell afterwards unless it was recorded at the time.

The honest limit is on the page too. The sweep stopped partway through when the device went unresponsive, so nine models still have no data. A full sweep costs roughly four minutes of measurement per model, plus the time to load it.

If you want your own numbers, TiinyBench is on the farm. By default it benchmarks whatever is already loaded and touches nothing else, the results are plain JSON, and the report is one HTML file you can open anywhere.

https://artifacts.semfreak.dev/a/tiiny/bench-cebc3011/

https://tiinyapp.farm/apps/tiiny-bench/

Post your numbers if you run it. I would like to see how the boxes compare.