ResearchBenchmarksScience Question Answering on ARC Challenge (Accuracy, Drop)Follow36.77AccuracyBF1636.686836.708436.7336.7516May 25, 2026Evaluation ResultsMethodMethodLinksAccuracyDropBF16shot-count=25-shot, Le...shot-count=25-shot, Learning Rate=10^-52026.0536.77—HiF8 QATshot-count=25-shot, Le...shot-count=25-shot, Learning Rate=10^-5, amax_algo=max, history=64, BF16 warmup=500 steps2026.0536.6922