“On the AA Briefcase benchmark, KIMI K3's performance was very close to Fable-5 and slightly ahead of GPT-5.6 Sol.”