ResearchBenchmarksZero-shot Reasoning on Reasoning Suite (PIQA, ARC, BoolQ, Winogrande)Follow83.68PIQA AccuracyFP1678.864880.114981.36582.6151May 21, 2026Evaluation ResultsMethodMethodLinksPIQA AccuracyARC-Easy AccuracyARC-Challenge AccuracyBoolQ AccuracyWinoGrande AccuracyAverage Zero-Shot Reasoning ScoreFP16#Bits=16-16#Bits=16-162026.0583.6883.4259.7385.0576.3277.64FP16*#Bits=16-16#Bits=16-162026.0583.4183.2955.884.5675.8576.58GEMQ#Bits=3-16#Bits=3-162026.0582.5480.6857.3485.0274.976.1GEMQ#Bits=2.5-16#Bits=2.5-162026.0581.0775.9351.7980.0973.7272.52EAQuant*#Bits=3-4#Bits=3-42026.0579.0578.4550.6878.6969.371.23