Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Over-Refusal Evaluation on OverBench

5.59ORR

w/o Safety-Aligned Optimization

4.558411.521718.48525.4483Jun 11, 2026
Updated 1mo ago

Evaluation Results

MethodLinks
2026.06
5.59
2026.06
6.64
2026.06
10.16
2026.06
12.69
2026.06
14.5
2026.06
15.86
2026.06
16.28
2026.06
26.17
2026.06
29.2
2026.06
30.99
2026.06
31.38