Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Safety alignment on SORRY-Bench (Score, log C)
Loading...
85.7
Score
πref
19.244
36.497
53.75
71.003
Jul 3, 2026
Score
Log C
Updated 18d ago
Evaluation Results
Method
Method
Links
Score
Log C
πref
Backbone=zephyr-7b, N=1
2026.07
85.7
5.44
πref
Backbone=zephyr-7b, N=16
2026.07
73.9
-
URIAL
Backbone=zephyr-7b, N=1
2026.07
63
3.15
URIAL
Backbone=zephyr-7b, N=16
2026.07
52.7
-
BoBN
Backbone=zephyr-7b, N=1
2026.07
43.9
-
BoBN
Backbone=zephyr-7b, N=4
2026.07
41.2
2.87
BoBN
Backbone=zephyr-7b, N=16
2026.07
36.1
-
πref
Backbone=tulu3-8b, N=1
2026.07
28.6
1.59
πref
Backbone=tulu3-8b, N=16
2026.07
26.8
-
URIAL
Backbone=tulu3-8b, N=1
2026.07
25.7
1.6
BoBN
Backbone=tulu3-8b, N=1
2026.07
24.3
-
BoBN
Backbone=tulu3-8b, N=4
2026.07
23.6
1.59
URIAL
Backbone=tulu3-8b, N=16
2026.07
22.6
-
BoBN
Backbone=tulu3-8b, N=16
2026.07
21.8
-
Feedback
Search any
task
Search any
task