BeaverTails
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Beavertails
92.03Safety Score
18
BeaverTails Text
99.3Overall Score
16
BeaverTails V
91.54F1 Score
15
BeaverTails (test)
98Blue Reward (Higher)
15
I-BeaverTails
14.68Attack Success Rate (ASR)
14
BeaverTails
64.5Attack Success Rate (ASR)
12
BeaverTails (BT)
99.1BT Score
12
BeaverTails Audio 1K
98.95RSR
12
BeaverTails
87.2Safety Rate
11
BeaverTails
69Reference Performance
10
Beavertails (test)
99.29Tracing Success Rate (TSR)
10
BeaverTails
77.9Det@1
8
BeaverTails
99.4Safety Score
8
BeaverTails (test)
84.3F1 Score
7
I-BeaverTails
12.58ASR
7
Beavertails
10ASR
7
BeaverTails-V Text-Image Prompt
88.36F1 Score
7
BeaverTails Harmful (held-out target labels)
0.793AUROC
6
BeaverTails V
63.44Accuracy
6
BeaverTails (test)
72.1Personalization Win Rate
6
BeaverTails (held-out)
76.7AUROC
5
BeaverTails
19.6ASR @ 1% FPR
5
BEAVERTAILS 1,000 examples native human labels (test)
85.5F1 Score
4
BeaverTails Self Harm
100Safe Rate
4
BeaverTails Child Abuse
90Safe Rate
4