Benchmarks & research
Open stack that turns Hugging Face LLMs into calibrated Jev-style System One models.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
uspraveen's Jevify combines the jev-bench dataset and leaderboard, an engine that reads answer probabilities instead of generating prose, training recipes and a TypeSafe-compatible /api/v1/systemone server. Its own models are distinct from the separately measured TypeSafe Jev 1.13.0 baseline. Reported accuracy and speed have not been independently reproduced. This is not altryne's Jevify or the pi-jevify extension.
Use jev-bench to compare a replacement model with the measured Jev baseline, rather than assuming the same answer format gives the same results. The maker's tables separate clear-cut questions from ones where people disagree, and show where Jev still leads.