Top AI Product

Fermat’s Last Theorem waited 358 years for a proof, then another 31 for a machine-checked one. Anthropic’s Claude just delivered it in 11 days.

What actually happened

This isn’t a product you can buy — it’s a research release, published September 4 with a public repo. An internal model “roughly equivalent to Claude Fable 5.1” ran as a swarm of agents on Prove2Me, a platform that keeps theorem statements in a dependency graph so agents prove pieces in parallel and reuse each other’s lemmas. The haul: 13 million lines of Lean code, 30,300 theorems, ~6 billion output tokens. That’s 5x the size of Mathlib — the library human mathematicians spent a decade building. Lean verified everything using only its three standard axioms. No gaps, no “sorry” placeholders.

Why it matters

Kevin Buzzard has led the human effort to formalize FLT since 2024, with a horizon stretching to 2029. His verdict on the AI version: autoformalization is now robust enough to build on. Humans gave only occasional high-level nudges. Next target is obvious — point this at the existing math literature and start catching the errors nobody noticed. HN agreed: 700+ points across two threads.

You Might Also Like

• 708 Github Stars in 48 Hours Claude Token Efficient Universal Claude md and the Fight Over Claudes Most Expensive Habit
• Anthropic Agent sdk Metered Credits Openclaw Reinstated June 15 Splits Your Claude Subscription in two
• Claude for Microsoft 365 ga Outlook Beta Anthropic Puts one Agent Inside Word Excel Powerpoint and Outlook
• Claude Cowork Cloud Anthropic Close Your Laptop the Agent Keeps Working
• Openwork Clones Claude Cowork as a Self Hosted Agent 97 Github Stars in a day