Harmlessness
Benchmarks
Dataset NameSOTA methodMetricTrendResultsLast Updated
96Average Win Rate
21
Apr 21, 2026
99.03Refusal Rate
20
Mar 4, 2026
99Refusal Rate
20
Mar 4, 2026
100Toxigen (%)
17
Feb 26, 2026
87.25Win Rate
6
Mar 5, 2026
89.99Win Rate
5
Mar 5, 2026
87.5Win Rate
5
Mar 5, 2026
2.772Reward
4
Feb 26, 2026