Claude formalizes Fermat's Last Theorem in 11 days with 13M lines of Lean code
新智元 · wechat · 2026-09-05
Anthropic says Claude completed the first end-to-end machine-verified proof of Fermat's Last Theorem in 11 days, working largely autonomously — generating 13 million lines of Lean code, 30,300 verifiable theorems (29,500 adopted), and burning 6 billion tokens, over 5x the size of Mathlib. The formalization was expected to take years.
Key details:
- Early multi-agent runs collapsed entirely (failed attempts = 7% of final code); the fix was Tianyi Peng's (Columbia/Anthropic, Tsinghua Yao class) Prove2Me platform: theorem-DAG task trees, statement/proof separation for faster compilation, and natural-language indexing for retrieval.
- Claude also proved 29,000 intermediate theorems across previously unformalized branches; Lean verified everything from just three base axioms.
- A side demo formalized Vinogradov's three-primes theorem in 3 days using 3 consumer accounts.
- Imperial College's Kevin Buzzard, who planned a years-long effort, called it a major step for automated formalization; Anthropic argues this ends the era of '99% certain' proofs.
Related event: Claude Produces First Formalized Proof of Fermat's Last Theorem in 11 Days(52 posts)→
More from Research
- A real larva's connectome — 2,675 cells, 14,066 connections — now swims in a 3D simulation — eptwts · 2026-10-21
- Stanford's T2-z lets robots probe hidden physics with a poke, lifting success from 34% to 73% — stanfordnlp · 2026-10-21
- Fei-Fei Li team unveils FACT, a force-aware foundation model for robot dexterity — stanfordnlp · 2026-10-21
- Upcoming paper claims big models can handle latency-critical real-time robotics — ManlingLi_ · 2026-10-21
- NVIDIA open-sources Flora, a robotics data engine 125x cheaper than GPT-6 Astra — DrJimFan · 2026-10-21
- openRxiv explains its approach to AI amid growing researcher adoption — anshulkundaje · 2026-10-21