Safety Benchmarks
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Safety Benchmarks (Sorry-bench, StrongREJECT, WildJailbreak, JBB-PAIR, JBB-GCG)
42.34Average Score
21
Safety Benchmarks ASB, InjecAgent, AHarm, ASecBench
61.1ASB Score
18
Safety Benchmarks VLG-1, VLG-2, Ch3EF
93.91VLG-1 Score
17
Safety Benchmarks (Final-day)
96.2Final-day Macro F1 Score
15
Safety Benchmarks (MMA, Sneaky, I2P-S, Ring, P4D)
80.4MMA
10
Safety Benchmarks
0.4Average ASR
10
Five Safety Benchmarks AdvBench, HarmBench, HarmfulQ, JBBench, StrongReject
7.69ASR
6
Safety Benchmarks Overall
0.001Cost per Accuracy Point ($)
4
Safety Benchmarks Aggregate (test)
73.6Generation Quality (Std Prefix)
4
Five Safety Benchmarks direct_q
0.02ASR
3