Claude computes nine-loop N=4 super-Yang-Mills amplitude
Anthropic says Claude met a physicist's public challenge by computing a nine-loop scattering amplitude in N=4 super-Yang-Mills on a modest budget, testing whether AI can get past known computational limits in theoretical physics.
Anthropic has published a guest post by former theoretical physicist and science writer Matt von Hippel. It describes how Claude answered a challenge he set for AI companies about a month earlier: to "take the kinds of computer resources an academic has access to" and solve a big open problem in scattering amplitudes.
Von Hippel offered two targets: N=8 supergravity to seven loops, or N=4 super-Yang-Mills to nine loops. Anthropic picked the second. According to the material, Claude (Fable 5.1, running in Claude Science) worked mostly on its own, spent a few thousand dollars, and produced a nine-loop result that was checked independently.
Scattering amplitudes let physicists predict how particles interact. Each extra "loop" makes the answer more precise and much harder to compute. Most amplitudes are known only to two loops. N=4 super-Yang-Mills is a "toy model," unrealistic because it has so many particles, and amplitudeologists use it to stress-test their methods.
Von Hippel says he chose a computational challenge on purpose. He wanted a problem that was out of reach because of limited compute and time, not because nobody knew how to do it.
Some caveats apply. Anthropic published and paid for the piece. The approach scales up an existing method rather than introducing a new one. A human team working with GPT-6 reportedly reached most of the same result at about the same time.
The source excerpt does not include technical details of the computation or how it was verified.