Claude wrote 13M lines of Lean to nail Fermat's Last Theorem

Anthropic's internal research model, Claude Fable 5.1, spun up dozens of agents and produced the first end-to-end, machine-checked proof of Fermat's Last Theorem in Lean 4: 13 million lines of code, 30,300 theorems proved, roughly 6 billion output tokens, in 11 days. The final artifact is about five times the size of Mathlib, the community library humans have been building for over a decade.

The unlock was coordination, not raw genius. A tool called Prove2Me held a directed acyclic graph of theorem statements, letting agents split the proof, claim nodes and check each other's work without collisions. Human input was thin — occasional high-level nudges from researcher Tianyi Peng. Lean signed off using only its three standard axioms, so nothing rests on trust.

Be precise about the claim: Claude did not prove FLT. Andrew Wiles did, in 1995. Claude translated a 129-page exposition of that proof into machine-verifiable form — autoformalization, not discovery. Mathematicians had penciled in several years for the job. Kevin Buzzard of Imperial College London called it a big step toward automatic formalization of the modern mathematical literature.

Why it matters: verification has been math's bottleneck — if agents can formalize a landmark proof in days instead of years, checking the work stops being the slow part.