“On the TerminalBench 2.1 benchmark, KIMI K3 scored 88.3, placing it just 0.5 points behind GPT-5.6-SOL and ahead of Fable-5.”