benchgap
Calibration

WildGuardTest (Decision Index) → HLE-MultiChoice (Decision Index)

HLE-MultiChoice (Decision Index) is estimated from WildGuardTest (Decision Index) with a linear curve fitted on 19 models measured on both: y = 0.5684·x + -0.3896, R² = 0.35, cross-validated error 2.6 pp. It is used for 0 estimates.

Fitted on

ModelWildGuardTest (Decision Index)HLE-MultiChoice (Decision Index)
Celeris-1.1 Decision94.9%12.5%
Celeris-1 Decision93.3%13.3%
Clef90.7%11.1%
Clef-Flash90.9%10.3%
Clef-Omni87.2%9.0%
d1-ultra95.4%15.0%
Drex Max95.5%16.0%
Drex Max v1.195.5%17.0%
Drex v1.586.9%12.1%
Drex v2.087.9%13.6%
elephant-193.3%18.3%
GLiDE95.1%10.5%
GLiDE v1.295.4%17.5%
Jev 1.1493.1%18.9%
d190.9%11.5%
Mercury Decide88.0%10.1%
Microsoft-Decision-188.2%10.9%
Perplexity Decider Flash v1.2 4B93.3%12.1%
Perplexity Decider v1.2 27B94.4%14.6%