Anthropic used Claude to compute a nine-loop amplitude in N=4 super-Yang-Mills. The result matched human work in progress by Song He’s group. Claude ran two methods: the bootstrap and a form-factor approach. It used 96 CPUs for a week at a cost of around one to two thousand dollars. The calculation was done via Claude Science with minimal human oversight. Anthropic physicists verified the result with Lance Dixon. The work shows AI can now handle frontier physics tasks autonomously. It also suggests more low-hanging fruit exists in computational physics. The field may need to redefine its frontier goals. The cost was feasible on academic-scale compute budgets.
What changed
Eight loops was the previous highest known amplitude in N=4 super-Yang-Mills.
What it unlocks
Computing a nine-loop scattering amplitude in a frontier theoretical physics problem with existing methods and modest compute.
- nine loops in N=4 super-Yang-Mills
- 96 CPUs for one week
- $1,000–$2,000 total compute cost
- two independent calculation methods
What you need to act on it
- Claude Science access
- Fable 5.1 model
Sources