jevals.com scores TypeSafe AI's Jev, the first System One model, and the LLMs that try to decide like it. Every model answers the same typed questions with probabilities and is graded against human labels: Decision Score, accuracy, calibration, cost and speed.
- Boards: noul (yes or no) · choice (pick one of N) · score (rubric)
- New to Jev? What is Jev? The System One model, explained
- Latest findings (release 2026-09-18): Jev ties the best of six LLMs on yes/no questions at 1/28 of the price. Read them
- Open data (CC BY 4.0): Jevals/jevals-data. Corrections and requests for a model row go in its issues.