Anthropic said on 25 September 2026 that Claude Fable 5.1, running autonomously inside its Claude Science platform, computed the six-particle scattering amplitude in N=4 super-Yang-Mills theory to nine loops — one loop past the eight-loop record Stanford's Lance Dixon set in 2023, and cheaply enough that the whole run fit on roughly 96 CPUs for about a week at total compute cost between $100 and $2,000.
Beating Dixon's record on a startup budget
Physicist Matt von Hippel issued the challenge in August 2026: point an AI at a frontier problem in amplitudeology, cap it at academic-scale hardware, and see whether it can do more than novelty demos. Fable 5.1 chose bootstrap methods, refined the ansatz over successive loop orders, and produced a result matching cross-checks against known limits. Dixon told Anthropic the feat is "quite a triumph, in my opinion, for a large language model to execute all of the steps."
Why the toy model matters
N=4 super-Yang-Mills is not the real world — it is the amplitudeologists' test bench, prized because its symmetries make higher-loop calculations tractable enough that new techniques can be tried out before being aimed at Standard Model physics. Each additional loop is exponentially harder; getting to nine, in a week, on a couple of thousand dollars of compute, is the kind of result that either changes the price of high-energy theory work or reprices it.
Fable and the Science stack
The run sits on top of Anthropic's newly launched Claude marketplace stack and the same Fable 5.1 line that Anthropic has been pitching for agentic scientific workflows. It also lands on a day when Anthropic said Claude autonomously identified a novel CRISPR-like enzyme system. Von Hippel's own quote — "There is more low-hanging fruit out there than you'd expect" — reads less like a boast and more like a research plan.
Reporting based on coverage from Anthropic, Unite.AI and AI Weekly.