whichjudge.dev

Terms

Last updated 2026-09-20. Contact: contact@radrebeldeveloper.com

1. This site is an independent measurement page. It is not affiliated with TypeSafe AI, OpenAI, or the dataset authors.
2. Numbers are from frozen public samples (seed=11, n=500 per task, except prompt injection at n=200 and review polarity at n=356, where the source rows left after the v1 run were fewer). They are not a warranty that a model will perform the same on your data.
3. Do not use these results as the sole basis for safety-critical routing (hate speech, medical, credit, employment).
4. You may cite the table with a link to whichjudge.dev and the run id shown on the page. Do not present it as an official TypeSafe benchmark.
5. The site is provided as-is. No SLA, no paid support unless we agree in writing.

Back