Disagreement Detection on CEFR-SP Qwen3 labels Rater A benchmark leaderboard · SOTA2 Research

Disagreement Detection on CEFR-SP Qwen3 labels Rater A

0.593ROC-AUCmost2

Evaluation Results

Method
most22026.05	0.593	0.591	90
logprob2026.05	0.567	0.539	55
multilingual-e5-large-instruct2026.05	0.567	0.568	95
deberta-v3-large2026.05	0.565	0.551	80
LaBSE2026.05	0.545	0.544	95
all-roberta-large-v12026.05	0.544	0.536	75
sentence-t5-base2026.05	0.543	0.537	60
instructor-large2026.05	0.542	0.539	85
bge-large-en-v1.52026.05	0.54	0.538	75
gte-large-en-v1.52026.05	0.54	0.538	80
sentence-t5-large2026.05	0.538	0.532	55
e5-large-v22026.05	0.534	0.537	80
exactw2026.05	0.525	0.508	60
exactw2026.05	0.482	0.509	75