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

General Capability on MTBench

9.14MTBench Score

REPBEND

Updated 1mo ago

Evaluation Results

Method	Links
REPBEND 2025.04		9.14
Original Weight Model 2025.04		8.71
WHP (W/O safe set) 2025.04		8.01
WHP (W/O safe set) 2025.04		7.88
Original 2025.04		7.84
NPO 2025.04		7.79
CB 2025.04		7.74
Original 2025.04		7.72
CB* 2025.04		7.72
NPO (W/O safe set) 2025.04		7.71
REPBEND 2025.04		7.71
APL (RM) 2025.05		7.65
WHP 2025.04		7.63
CAPO 2025.05		7.63
APL 2025.05		7.59
CB 2025.04		7.58
DPO 2025.04		7.57
CB* 2025.04		7.57
APL (w/o Adv) 2025.05		7.56
DPO 2025.04		7.54
DPO 2025.05		7.54
Base 2025.05		7.53
Safety Prompting 2025.04		7.5
REPBEND 2025.04		7.5
REPBEND 2025.04		7.37
Original Weight Model 2025.04		7.35
NPO 2025.04		7.31
Safety Prompting 2025.04		7.18
TA 2025.04		7
TA 2025.04		6.93
SFT 2025.04		6.88
APL (RM) 2025.05		6.88
DPO 2025.05		6.85
Base 2025.05		6.78
CAPO 2025.05		6.66
APL 2025.05		6.59
APL (w/o Adv) 2025.05		6.53
WHP 2025.04		6.29
SFT 2025.04		5.98
R2D2* 2025.04		5.97
RMU 2025.04		5.77
NPO (W/O safe set) 2025.04		4.67
RMU 2025.04		4.01