What it does
Its headline finding is that Jev ties the best of the six on the yes/no task at roughly a twenty-eighth of the price, and that no model clearly beats guessing on the preference-scoring task.
Maker-reported (not independently measured by JevMade): Jev Decision Score: 69.0 (noul, PubMedQA), 67.8 (choice, Banking77), 9.2 (score, HelpSteer2) · Jev costs $0.029 per 1,000 decisions, p95 653 ms (noul); $0.043, p95 693 ms (choice); $0.036, p95 670 ms (score) · 8 systems, suite 0.1.0, release 2026-09-18
- Primitives
- choice, score, noul
- Added
- Project created
Source checked 2026-09-24 — opened the GitHub repository directly.