An AI model developed by Anthropic solved a complex scattering amplitude problem that had remained out of reach for theoretical physicists for years. The tool, known as Claude Science, successfully computed nine loops in the N=4 super-Yang-Mills model, surpassing the previous eight-loop limit established by researchers at SLAC. While most particle physicists stop at three loops because the required computing power grows exponentially, the AI operated largely unsupervised for several days to complete the task.

Professor Lance Dixon of SLAC and Stanford independently verified the result and noted that the AI's grasp of relevant 2019 and 2023 research papers exceeded that of most humans. The accomplishment followed a challenge issued on August 7 by physicist Matt von Hippel, which Claude finished within 25 days at a cost of roughly $1,000 to $2,000. Other teams, including a group in Beijing using GPT-6, had reached partial results, but Anthropic's tool solved the problem in one shot, scooping human teams that had spent years on the calculation.