01 What happened
Anthropic reports that Claude worked largely autonomously over 11 days to write a complete formal proof of Fermat's Last Theorem in the Lean programming language. The company says the work produced 13 million lines of Lean and 29,500 intermediate theorems. The proof is public on GitHub for outside checking, and all figures come from Anthropic's own report.
02 Key details
- Anthropic says the proof uses only Lean's three standard axioms, and a comparator confirmed the theorem's statement matches Mathlib's own statement of FLT.
- Anthropic says human input was limited to occasional high-level instructions from Tianyi Peng, an Anthropic researcher. Some early attempts by Claude failed, and success followed a switch to the Prove2Me platform.
- Anthropic says failed early efforts contributed about 7% of the non-boilerplate lines in the final proof.
- The effort consumed about six billion output tokens from an internal general-purpose research model, which Anthropic says is roughly comparable to Claude Fable 5.1.
03 Why it matters
If the claims hold, they suggest AI agents can produce very large machine-checked formal proofs in days rather than years. That could make it easier to verify new mathematical results, including AI-generated ones. The result rests on Anthropic's own report, so outside review of the public proof remains the key test.
04 Who it matters to
Mathematicians, formal verification specialists and artificial intelligence researchers.