AI Model Claude Unravels 387-Year-Old Mathematical Enigma in Just 11 Days

In a monumental accomplishment for both artificial intelligence and the field of mathematics, Anthropic declared on September 5, 2026, that its AI model, Claude, has successfully generated the world’s first comprehensive, computer-verified proof of Fermat’s Last Theorem. This theorem is one of the most renowned unsolved problems in mathematical history, initially proposed by Pierre de Fermat in 1637.

Operating largely independently over a span of 11 days, Claude formalized the proof using Lean, a dedicated mathematical programming language that enables computers to automatically scrutinize logical proofs for potential errors. The outcome was a staggering 13 million lines of Lean code and 29,500 intermediate theorems proved in the process, making it the most extensive Lean proof ever composed.

The initial proof of Fermat’s Last Theorem was formulated by mathematician Sir Andrew Wiles in 1995, spanning across 129 pages. Experts had projected that formally verifying and computerizing Wiles’ proof would require a team of specialist mathematicians several years. However, Claude, utilizing a multi-agent system powered by the open-source tool Prove2Me, achieved this feat in less than two weeks, consuming approximately 6 billion output tokens.

This breakthrough underscores the escalating capability of AI to execute rigorous, prolonged mathematical reasoning, and it could fundamentally revolutionize the way researchers validate complex proofs in the future.

Source: Anthropic Research – Formalizing Fermat’s Last Theorem