Public Benchmarks
Four independent-seed blind rounds + three-model shared-board measurement: 97.45% mean, 17-category details, three-layer OOD disclosure, and honest weaknesses — all public.
97.45%
4-round blind-test mean (±0.6pp)
17
categories covered (12 perfect)
2,000
real calls, zero errors
100%
data & scripts published
| Metric Scope | amos-2.8 | jev-1.13 (commercial closed-source) | laya (open-source lightweight model) |
|---|---|---|---|
| Overall accuracy (500 questions) | 96.6% | 96.4% | 42.8% |
| Tier A pure-decision tasks (280 questions) | 96.05% | 95.81% | 37.0% |