Kimi K3's reported benchmark gap with GPT-5.6 Sol and Claude Fable 5

A user says their analysis of more than 90 benchmarks released after Kimi K3 puts it close to the older GPT-5.5 and Opus 4.8, but behind GPT-5.6 Sol and Claude Fable 5. K3 lost all 10 agentic-coding benchmarks they tracked against Fable. They say it’s too early to tell whether the broader gap between open-source and proprietary models has narrowed.