Anthropic says Claude solved a frontier physics calculation

Anthropic published a guest post saying Claude Science completed a frontier theoretical-physics calculation: a nine-loop scattering amplitude in N=4 super-Yang-Mills, a toy model physicists use to stress-test methods for particle-physics predictions.

The challenge came from physicist and science writer Matt von Hippel, who asked AI companies to show whether a model could push past what looked like a practical compute limit in his former field. Anthropic researchers used Claude Science with Fable 5.1 and say the system produced the result through two approaches, including a bootstrap method and an indirect form-factor method.

The post says the work was checked with Lance Dixon of SLAC, and that a separate human group had also made progress with AI assistance. Anthropic frames the result carefully: Claude did not invent a magic new physics method, but it handled a long, finicky calculation with modest compute and little human steering.

That is the part that matters outside particle physics. If the report holds up through the researchers’ own publications, it is another sign that AI research tools are moving from solving tidy benchmark problems toward carrying extended scientific workflows where reliability, software discipline and persistence matter as much as raw reasoning.