Moonshot released Kimi K3 July 17: 2.8 trillion parameters, fully open-source. Artificial Analysis independently ranks it ahead of Anthropic's Opus 4.8 on frontier benchmarks, first Chinese open-weight model to do that. Still behind Claude Fable 5 and GPT-5.6 overall, but Moonshot doesn't claim otherwise. Artificial Analysis and Arena.ai placed it there independently. It also topped web interface engineering evals in blind human-preference comparisons against Claude Fable. Three competing Chin...