Kev 4B Release - JevBench Public All 231 Accuracy: leaderboard

231 / 111 Two boards (all 231; hard 111). Do not merge into "JevalKit - JevBench" (Kev 4B 74.03 there is a different evaluator run). ECE 0.112 held (panel ambiguous). Metric: accuracy.

Metric: accuracy. Source: huggingface.co. 1 model tracked. Updated 14 Nov 2026.

Top models

#ModelScoreOverall rank
1Kev 4B0.758

Interactive version: theaggregate.ai/benchmark?slug=kev-4b-release-jevbench-public-all-231-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-16.