In a post on X, Haven Feng introduced Schema, which he described as a harness that reaches 99% RHAE on the ARC-AGI-3 public set with Opus 4.8 plus Fable 5 and 95.35% with GPT-5.6 Sol, while making an LLM "think like a physicist." Naval amplified it as work from Impossible Research, calling it an agent harness that plays games, writes code, reasons like a physicist and "saturates" the benchmark. So far, the public evidence is the company's own announcement and the discussion around it on X.