“On an internal frontend evaluation benchmark, KIMI K3's performance is closest to Opus 4.7, placing it approximately 3 months behind frontier models like GPT-5.6 SOL and Fable-5.”