List your company

Confidence Calibration on GPQA (test)

0.102ECEHTC
0.087920.182960.2780.37304Jan 22, 2026

Evaluation results

MethodLinks
HTCVariant=Reduced2026.010.1020.2130.706
HTCVariant=Full2026.010.1240.2190.704
GlobalTrace-TPTemperature Scaling=true2026.010.1360.2570.643
LastStep-TPTemperature Scaling=true2026.010.1390.2580.61
GlobalTrace-TPTemperature Scaling=false2026.010.4140.4020.649
LastStep-TPTemperature Scaling=false2026.010.4240.4130.614
Verbalized Conf2026.010.4540.5230.593