Meta Releases AutoformBot for Large-Scale Formal Mathematics
- •Meta AI launched AutoformBot, a multi-agent framework to formalize mathematical textbook prose into machine-checked code.
- •The project created Atlas, a library containing over 45,000 Lean 4 declarations and 500,000 lines of code.
- •AutoformBot successfully processed a corpus of 26 open-access graduate-level mathematics textbooks across various fields.
Meta AI researchers introduced AutoformBot on May 27, 2026, as a multi-agent system designed to convert informal mathematical textbook prose into machine-verifiable code. The framework operates by coordinating thousands of LLM agents equipped with formal verification tools, dependency-aware task scheduling, and collaborative version control to ensure accuracy.
The team successfully applied this method to a corpus of 26 open-access textbooks covering subjects including analysis, algebra, topology, combinatorics, and probability. This process generated Atlas, a library containing over 45,000 Lean 4 declarations and 500,000 lines of code. Lean 4 is a functional programming language used for building formal mathematical proofs.
Meta has released both the AutoformBot framework and the Atlas library as open-source assets. The project demonstrates that autoformalizing graduate-level mathematics at scale is technically and economically feasible, potentially enabling future automated verification of research-level mathematical content.