Safety Alignment Evaluation
Benchmarks
Dataset NameSOTA methodMetricTrendResultsLast Updated
82.14Harmfulness (%)
36
Feb 26, 2026
90.23Win Rate (WR)
32
Mar 10, 2026
0.7Harmful Response Rate
18
May 27, 2026
4.2Harmful Response Rate (%)
18
May 27, 2026
99Refusal Rate
16
Feb 26, 2026
100Safety Score
14
Jun 4, 2026
93.33Safety Score
14
Jun 4, 2026
85.03Safety Score
14
Jun 4, 2026
98.6Safety Rate
14
Jun 2, 2026
99Safety Rate
14
Jun 2, 2026
100Safety Rate
14
Jun 2, 2026
100Safety Rate
14
Jun 2, 2026
100Safety Rate
14
Jun 2, 2026
98.72Safety Rate
14
Jun 2, 2026
100Safety Rate
14
Jun 2, 2026
0.5Average Absolute Improvement
12
May 27, 2026
53Utility
9
May 15, 2026
54Utility
9
May 15, 2026
14Harm Score (HS)
7
Jun 11, 2026
37.2HS Score
7
Jun 11, 2026
1.6Harm Score (HS)
7
Jun 11, 2026
4Harmful Score (HS)
7
Jun 11, 2026
81Response Safe Rate (Llama Guard Model)
5
Mar 5, 2026
1.3MMA
3
Jul 2, 2026
39.1JSR
3
Jun 4, 2026