Fable 5 closes 81.7% of the human AI-research gap

Prime Intellect ran 153 autonomous experiments across 18 frontier models, and Claude Fable 5 closed 81.7% of the gap between a 3,290-step baseline and a 2,600-step human record on its nanoGPT optimizer speedrun. The public leaderboard shows Opus 5 at 53.6% and GPT-5.6 Sol at 35.9%; Prime Intellect’s methodology says runs lasted up to eight days on 8×H200 nodes and produced no fundamentally new method. Research-agent teams should treat experiment selection and noise handling—not a single leaderboard score—as the capability to measure, extending the archive’s reliability work on multi-agent research loops.