GLM-5.2 took first on Design Arena, leapfrogging a “now unavailable” Claude Fable 5 — a phrase doing real geopolitical work. Anthropic’s consolation was Epoch’s Capabilities Index, where Fable 5 beat GPT-5.5 Pro by a point, its first lead in a year, and reaches almost no one.
An AI chemist ran 10,080 experiments in a robotic lab and taught itself the trick that made a stubborn reaction work.