yev0 Author Benchmark - R-Judge571: leaderboard

yev0 Author Benchmark - R-Judge571 — Released lc100 checkpoint, BF16 adapter onbase, transformers calibrated letter-logit readout with exact system/JSON prompts and enable_thinking=False. Native author run, not copied peer comparison. DynaBench FAIL positive; R-Judge unsafe positive. Train overlap audit in primary card; no official multi-axis JevBench score claimed. Metric: F1 (%). Source: https://huggingface.co/choyiny/yev0-4b

Metric: F1 (%). Source: huggingface.co. 1 model tracked. Updated 16 Nov 2026.

Top models

#ModelScoreOverall rank
1yev0-4B50.8

Interactive version: theaggregate.ai/benchmark?slug=yev0-author-benchmark-r-judge571 · How It Works · Data refreshed daily, snapshot 2026-11-16.