The A.I. Daily
Four open decision models benchmarked on an RTX 4090
A user benchmarked four new open decision models, Laya, d1 3B, Clef-Flash 9B and Lev 4B, on a single RTX 4090, tasking them to flag centipede names word by word across 9,534 words of Wikipedia text. Laya was fastest overall at 3.9 ms per word, about 7,980 words in 32 seconds and 97.4% accuracy; Lev was slowest but caught the most names.