Claude Breaks Nine-Loop Physics Record, Scooping Human Team

Anthropic's science blog post "Yes, Claude can do Nine Loops" announced that Claude, responding to a challenge issued last month by physicist Matt von Hippel (4gravitons), computed a nine-loop six-particle scattering amplitude in planar N=4 super-Yang-Mills, breaking the eight-loop record set in 2023 by Lance Dixon's team at SLAC. It is widely seen as a landmark case of AI achieving a frontier theoretical-physics computation, though the challenger's own assessment and human researchers being scooped urge caution about its significance.

Confirmed

• The challenge came from 4gravitons last month; within a month Claude pushed the N=4 SYM amplitude to nine loops, announced on Anthropic's official blog. Per @EricBuess, it was a guest article written by 4gravitons, and physicist Lance Dixon personally verified the result.
• Per multiple accounts (@DrSingularity, details cited by Kyle Cranmer), Claude ran essentially unsupervised for days while researchers mostly said "keep going"; it wrote its own code from scratch using established physics methods, computed the amplitude two independent ways, and all 107,053 nonzero coefficients matched exactly, reportedly in one pass, at roughly hundreds of dollars per method; one account (@minchoi) puts total cost at 1000–2000 USD, a slightly different figure.
• Scattering amplitudes are core formulas for predicting particle behavior and are notoriously hard; most research stops at two or three loops, with cost growing exponentially per loop. N=4 SYM is regarded as the "perfect sphere" of QFT and a common testing ground.
• Physicist Kyle Cranmer said his team's work was scooped by Anthropic; Cambridge physicist Ben Allanach revealed his team, including collaborator Lance Dixon, was scooped too. Dixon wrote "What it feels like to be scooped by a machine?" noting in an appendix that his method's steps closely mirror Claude's.

Unconfirmed

• 4gravitons' own assessment: this is not millennium-prize level—the problem has a well-polished recipe across multiple papers, and he expected it would need novel techniques but little compute, which proved true; some in the physics community (e.g., @burnytech) argue a true milestone would be closer to recognized open problems.
• Per @burnytech relaying 4gravitons, another team obtained most of the result with less AI assistance in the same period; he also disclosed being invited to write for Anthropic.
• A single post (@1411337) mentions a specific version/platform (Claude "Fable 5.1"), unverified by official sources.

Why it matters

• It is cited as evidence of AI's capacity for long-horizon autonomous research: days of unsupervised work, code from scratch, two-method cross-validation passing in one shot at costs from hundreds to low thousands of dollars.
• Yet 4gravitons' caveat tempers the narrative: the problem followed a mature recipe, and non-AI teams were close to done; the achievement is more engineering execution than new physics.
• For physics, being "scooped by AI" has sparked discussion about research pace and human-AI collaboration, as reactions from Cranmer, Allanach and Dixon show.

2026-09-26 ~ 2026-09-27 · 27 related posts

• [source] Anthropic: Claude solves nine-loop scattering amplitudes, breaking the eight-loop record — AnthropicAI · 2026-09-26
• Anthropic says Claude can compute notoriously hard Nine Loops physics amplitudes — daniel_mac8 · 2026-09-26
• [source] Scooped by a Machine: Claude Computes N=4 Super Yang-Mills to Nine Loops — KyleCranmer · 2026-09-26
• Claude one-shots a nine-loop scattering amplitude, a frontier physics calculation — Dr_Singularity · 2026-09-26
• Physicists find Anthropic's AI scooped their research result; collaborator checked it under NDA — benitoz · 2026-09-26
• Physicist assesses Claude's nine-loop scattering amplitudes: easier than a Millennium problem, but still surprising — burny_tech · 2026-09-26
• One Prompt, Days of Autonomous Running: Claude Solves Nine-Loop Amplitude for ~$100 of Compute — AI寒武纪 · 2026-09-26
• Claude computes nine-loop scattering amplitudes, physicists ask: where's the novel theory? — burny_tech · 2026-09-26
• Claude pushes physics calc to 9 loops, physicist Cranmer: frame it as human-AI contribution — KyleCranmer · 2026-09-26
• [source] Physicist says Anthropic's Claude scooped his team, mimicking their methods and format — KyleCranmer · 2026-09-26
• Claude Fable 5.1 Completes Frontier Nine-Loop Particle Physics Calculation Experts Chased for Years — 141_1337 · 2026-09-26
• Claude computes nine-loop scattering amplitude, verified by SLAC's Lance Dixon for ~$1–2K — EricBuess · 2026-09-26

15 near-duplicate retellings: Dr_Singularity · rohanpaul_ai · kevinweil · ctjlewis · ctjlewis · burny_tech · AlexTensor · 机器之心 · 新智元 · 智东西 · rohanpaul_ai · CurieuxExplorer · LucaAmb · minchoi · burny_tech