“On major proprietary benchmarks, Kimi K3's performance typically matches or slightly trails that of Claude Fable 5 and GPT 5.6 Sol.”