Briefs · Emerging Technologies
China just dropped a new frontier model... and nobody can explain how
China’s frontier model should be impossible. Kimi K3.1 is a free, 3-trillion-parameter open-weight model from Moonshot AI, released last Friday after masquerading as the mystery model Kibitz on Chapot Arena. It sits within one point of Opus 5.5 on Terminal-Bench 4.0, and Artificial Analysis independently scored it 67%, the highest of any model tested. The paradox is that Big AI insists Chinese models only improve by stealing outputs, yet Anthropic also claims it now blocks 90–99% of distillation attacks. The verdict is hilariously horsey: Kimi built the most beautiful 3D horse on Thoroughbench, then animated it like a wobbling jelly.
Read the original at Fireship (The Code Report)
Summary in our own words. The original belongs to its publisher.