What it does
It includes a Jev benchmark comparison and a WebGPU playground that runs in the browser.
Benchmarks & research
Verdict is a 151M-parameter ModernBERT decision engine with calibrated uncertainty and no autoregressive generation.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
It includes a Jev benchmark comparison and a WebGPU playground that runs in the browser.
Your app can read a message and pick the right answer from a list. A developer can try this in a web browser using the included test page. The tool rates how sure it is about its choice. Your app can then wait for a person to decide if the answer is unclear.
The tool can look at up to 24 choices at one time. Giving it more options will cause an error. It can also get confused if the right answer is missing and the other choices look similar.