Claude does nine loops, one past the N=4 physics record

Given a single prompt, Claude ran largely unsupervised for days in Claude Science and solved a nine-loop scattering amplitude for a few thousand dollars. The previous record in planar N=4 super-Yang-Mills, a simplified model physicists use as a testing ground, was eight loops, set by SLAC's Lance Dixon and collaborators.

Scattering amplitudes predict how particles behave. Each added loop of corrections sharpens the answer but takes exponentially more computation, so most calculations stop at two or three.

The run answered a challenge @4gravitons posed last month: get past eight loops on a compute budget an academic could reasonably access. Claude used methods from Dixon's group, and Dixon verified the result independently.

Anthropic doesn't give an exact cost or say where humans stepped in during the run.