校准置信度
The degree to which a model's stated confidence matches its actual accuracy.
Artificial Analysis launched Decision Index v2, raising private dataset coverage from 25% to 45% and adopting a headline metric that rewards both correctness and calibrated confidence. Nace AI's Drex Max v1.1 narrowly leads at 73, while Mammoth 27B lands within 4 points of GPT-6 Luna at 118x the speed.
Artificial Analysis launched Decision Index v2, raising private-dataset coverage from 25% to 45% and introducing a Decision Score that rewards both correctness and calibrated confidence. Drex Max v1.1 leads narrowly at 73, just ahead of GLiDE v1.2 and Mammoth 27B, while GPT-6 Luna (Max) scores higher but costs $1,530 per 1M tasks and runs ~40x slower.