Anthropic says Claude can calculate nine-loop particle-scattering formulas
The post points to theoretical physics, where loop order measures successive corrections; the number alone does not show what Claude calculated or how it was checked.
By Ryan Merket · Published
Primary source: X
Why it matters
Anthropic is positioning Claude for scientific work, where a high loop count is meaningful only alongside the specific calculation and its verification. The announcement puts a technical capability claim on the record without, by itself, establishing a new physics discovery or general performance across problems.
Anthropic said on September 25th that Claude can calculate particle-scattering amplitudes through nine loops, a claim the company introduced in a post on X announcing new work on its Science Blog.
The post describes scattering amplitudes as formulas theoretical physicists use to predict how particles behave. It calls the calculations notoriously hard and says researchers build up increasingly fine corrections, or "loops." In quantum field theory, loop order refers to successive contributions in a perturbative calculation. It is a measure of how far a calculation extends, not a count of particles, experiments, or repeated runs of Claude. A review of multi-loop Feynman integrals describes these calculations as part of higher-order predictions in the Standard Model and effective field theories.
The nine-loop figure is striking, but it does not stand on its own as a measure of what the model accomplished. A result at that order can depend heavily on the theory, the particular amplitude, the available mathematical structure, and the amount of human guidance. Showing that a model can reproduce a known expression would demonstrate a different capability from deriving a new result, and both differ from producing a result that researchers independently verify. The announcement's loop count does not distinguish among those tasks.
Nine-loop amplitudes also predate this Claude claim. A 2016 physics paper reported the parity-even part of a five-point amplitude through nine loops in planar N=4 supersymmetric Yang-Mills theory, alongside related calculations at higher orders. That provides useful context, not a direct comparison: Anthropic's post does not identify the calculation or theory needed to determine whether the work matches that earlier result or tackles a different problem. The historic result is documented in the paper's abstract.
For physics, higher-order calculations can improve theoretical predictions and help researchers interpret measurements. The mathematical workload is formidable: multi-loop work may involve complicated integrals and symbolic expressions, and physicists use methods that exploit the structure of amplitudes rather than treating every calculation as a brute-force expansion. The SAGEX review surveys the role of analytic integration and simplification in these calculations. That context makes the claim relevant to scientific computing, while also explaining why a single headline number cannot establish the breadth of the capability.
Anthropic's Science page presents the work as part of a program exploring AI's use in scientific research. The strategic proposition is that Claude can contribute to technical work beyond routine coding and prose: symbolic calculations in fields where progress often depends on manipulating dense mathematical objects. The claim would carry more weight as an advance in research if its scope and verification could be judged alongside the loop count. As presented in the X post, nine loops is Anthropic's headline for the result; the number alone does not establish that Claude independently discovered a new physical result or that the method generalizes across scattering problems.