“On GDPval's test of economically valuable tasks, Mythos/Fable-5 scored 1932, outperforming both Opus 4.8 (1890) and GPT-5.5 (1769).”