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

Unsafe Robustness on HarmBench

1Unsafe Rate

Dream-Instruct

-2.16819.21640.661.984Apr 28, 2026
Updated 22d ago

Evaluation Results

MethodLinks
2026.04
173
2026.04
1.240.5
2026.04
1.867
2026.04
1.873
2026.04
2.867
2026.04
340.5
2026.04
3.867
2026.04
473
2026.04
4.240.5
2026.04
4.240.5
2026.04
4.567
2026.04
573
2026.04
6.269.2
2026.04
7.269.2
2026.04
7.569.2
2026.04
7.569.2
2026.04
8.55
2026.04
96
2026.04
9.25
2026.04
9.80.2
2026.04
10.56
2026.04
10.56
2026.04
10.50
2026.04
115
2026.04
11.86
2026.04
11.80.2
2026.04
11.85
2026.04
12.90
2026.04
19.535.5
2026.04
25.839
2026.04
3439
2026.04
34.535.5
2026.04
35.535.5
2026.04
3635.5
2026.04
37.239
2026.04
3839
2026.04
46.50.5
2026.04
470
2026.04
54.80
2026.04
54.80.2
2026.04
56.50.8
2026.04
570.2
2026.04
57.50.2
2026.04
57.80
2026.04
580.2
2026.04
580.2
2026.04
580
2026.04
58.20
2026.04
58.50
2026.04
58.50
2026.04
58.80
2026.04
58.80.2
2026.04
58.80
2026.04
59.80
2026.04
59.80
2026.04
600
2026.04
60.20
2026.04
60.50
2026.04
60.50
2026.04
620
2026.04
68.20.5
2026.04
69.80.5
2026.04
70.50.8
2026.04
70.50.5
2026.04
71.50
2026.04
72.50.8
2026.04
74.80
2026.04
75.50
2026.04
77.50
2026.04
78.50
2026.04
79.50.8
2026.04
80.20