Claude formalizes Fermat's Last Theorem in 11 days with 13M lines of Lean code
AI寒武纪 · wechat · 2026-09-05
A team led by Anthropic researcher Tianyi Peng had Claude nearly autonomously produce a complete Lean machine-verified proof of Fermat's Last Theorem in 11 days — over 13 million lines of code (5x the size of Mathlib), with 29,500 of 30,300 attempted derived lemmas accepted and 6B output tokens consumed. Blog and code are public.
The key enabler was Columbia's Prove2Me collaborative platform: a DAG of theorems to preserve global context across parallel agents, separate files for statements and proofs to speed compilation, and natural-language descriptions for retrieval. Human input was minimal. Kevin Buzzard called the automated formalization, resting only on Lean's axioms, a major step toward machine-verifying all of modern mathematics. Using three regular Claude Max subscriptions, the team also formalized Vinogradov's three-primes theorem in 3 days.
Related event: Claude Produces First Formalized Proof of Fermat's Last Theorem in 11 Days(52 posts)→
More from Research
- O'Reilly builds a working data vocabulary for the semantic era, from warehouses to ontologies — rseroter · 2026-09-22
- Glance reads structured visual answers from a frozen 4B VLM, cutting GPU cost up to 85% — multiply_matrix · 2026-09-22
- SkillLift cuts agent skill-evolution token cost 40-70% by ranking, not rollouts — dair_ai · 2026-09-22
- FAIR & NYU scaling law paper pushes fits down to 4M-param models — giffmana · 2026-09-22
- Salesforce: Fine-tuning a weak model to copy Gemini drops success 15%; correcting its own failures works — rohanpaul_ai · 2026-09-22
- Jevons effect sparks encoder model revival: 'BERT revenge' over GPT — JFPuget · 2026-09-22