Anthropic formalizes Fermat's Last Theorem in Lean with 13.4M-line proof, completing 100-theorem benchmark
AlexKontorovich · x · 2026-09-05
Anthropic announced that one of its internal models, using the prove2.me platform, has produced a complete Lean formalization of Fermat's Last Theorem.
Key facts:
- The proof spans over 13.4 million lines of code, more than 5x Mathlib, and compiles roughly 20x slower than Lean's math library even on a 96-core, 500GB-RAM machine
- It follows the Darmon–Diamond–Taylor 1995 exposition of the Wiles–Taylor–Wiles argument, via Langlands–Tunnell and Ribet's level-lowering, developing Fontaine theory and enough of Mazur's Eisenstein ideal work to rule out Frey curves with a point of a certain order
- This closes out the last item in Freek Wiedijk's famous list of 100 formalization challenges, a 20-year benchmark
Kevin Buzzard and the Xena project blog independently compiled and verified the code with the comparator — it checks out.
Related event: Claude Produces First Formalized Proof of Fermat's Last Theorem in 11 Days(52 posts)→
More from Models
- Opus 5.5 tops OpenRouter spending in 20 of 29 task categories — mmurph · 2026-10-21
- Meta team builds open-source 30B LLM from scratch in months, hails 'second era' — TimDarcet · 2026-10-21
- DFlash drafter for Muse Glimmer-30B: 5 layers, blockwise 16-token spec decoding — TimDarcet · 2026-10-21
- EvoPress quantization shrinks 30B model from 56GB to 19GB at near-bf16 accuracy — TimDarcet · 2026-10-21
- Radar chart: Muse Glimmer-30B benchmarked against Gemma4-31B and Qwen3.6-27B — TimDarcet · 2026-10-21
- Privacy evals: Muse Glimmer-30B leaks less than Qwen3.6-27B on CIMemories — TimDarcet · 2026-10-21