Anthropic’s Claude AI has generated a computer-verified proof of Fermat’s Last Theorem in 11 days. The resulting document consists of 13 million lines of code, making it the longest mathematical proof ever created.
The project was led by Tianyi Peng from Columbia University, who utilized a team of Claude agents working in parallel. The agents used a tool called Prove2Me to coordinate tasks and prevent redundant work, though approximately 7% of the final lines resulted from initial false starts.
The AI proved over 30,000 supporting theorems during the process, consuming billions of tokens. The system operated on a research model comparable to the public Claude Fable 5.1, and the final output is five times larger than the existing Mathlib library.
Mathematician Kevin Buzzard, who is leading a separate human-led project at Imperial College London to formalize the same theorem, reviewed the AI’s work. He confirmed that the proof is valid and relies solely on basic logical axioms.
While the original proof of Fermat’s Last Theorem was established by Andrew Wiles in 1995, the AI’s contribution provides a formal, machine-checkable verification. The full 13-million-line proof is currently available on GitHub for public review.