---
format: "aidr-story-markdown/v1"
id: "c918d9184357963a6fd740e7e17a7cabf9dec2eeacbcce7d59935dbb83dd2a40"
canonical_url: "https://aidr.today/c918d918?lang=en"
title: "Claude Produces First 13 Million Line Lean Proof of Fermat's Last Theorem in 11 Days"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-09-04T20:52:10.000Z"
category: "Research"
topics: ["anthropic","claude","multi-agent","reasoning","coding","open-source","agent"]
source_urls: ["https://huggingnews.com/ai/claude-produces-first-13-million-line-lean-proof-of-fermats-last-theorem-900c9034","https://x.com/AnthropicAI/status/2095947707605266436","https://x.com/scaling01/status/2095953401460768990","https://x.com/sammcallister/status/2095950711380910526","https://x.com/rohanpaul_ai/status/2095965328669053054","https://lobste.rs/s/ccl4ag/formalizing_fermat_s_last_theorem","https://www.anthropic.com/research/formalizing-fermats-last-theorem","https://x.com/BlackHC/status/2096504623544639516"]
summary: "Dozens of agents used a process called autoformalization to turn advanced human mathematics into proofs that software can check line by line. The resulting output verifies a theorem conjectured over 350 years ago and first proven by Sir Andrew Wiles in 1995. Anthropic's Claude AI completed the formalization in 11 days, producing a 13 million line Lean proof, which is the largest ever written, including 29,500 intermediate theorems across geometry and number theory. The company stated the project marks a step toward firming up the core of mathematical knowledge and reducing the labor involved in refereeing proofs. Researchers said the system proves that AI autoformalization artifacts are robust enough to be built upon, allowing for the verification of complex, multi layered reasoning without assumptions beyond basic axioms."
---

# Claude Produces First 13 Million Line Lean Proof of Fermat's Last Theorem in 11 Days

> [Open the canonical story](<https://aidr.today/c918d918?lang=en>)

**Published:** 2026-09-04T20:52:10.000Z
**Category:** Research
**Topics:** anthropic, claude, multi\-agent, reasoning, coding, open\-source, agent

## Summary

Dozens of agents used a process called autoformalization to turn advanced human mathematics into proofs that software can check line by line\. The resulting output verifies a theorem conjectured over 350 years ago and first proven by Sir Andrew Wiles in 1995\. Anthropic's Claude AI completed the formalization in 11 days, producing a 13 million line Lean proof, which is the largest ever written, including 29,500 intermediate theorems across geometry and number theory\. The company stated the project marks a step toward firming up the core of mathematical knowledge and reducing the labor involved in refereeing proofs\. Researchers said the system proves that AI autoformalization artifacts are robust enough to be built upon, allowing for the verification of complex, multi layered reasoning without assumptions beyond basic axioms\.

## Sources

- [Story source](<https://huggingnews.com/ai/claude-produces-first-13-million-line-lean-proof-of-fermats-last-theorem-900c9034>)
- [Story source](<https://x.com/AnthropicAI/status/2095947707605266436>)
- [Supporting source](<https://x.com/scaling01/status/2095953401460768990>)
- [Supporting source](<https://x.com/sammcallister/status/2095950711380910526>)
- [Supporting source](<https://x.com/rohanpaul_ai/status/2095965328669053054>)
- [Discussion](<https://lobste.rs/s/ccl4ag/formalizing_fermat_s_last_theorem>)
- [Story source](<https://www.anthropic.com/research/formalizing-fermats-last-theorem>)
- [Story source](<https://x.com/BlackHC/status/2096504623544639516>)