formal-proof

Anthropic's unreleased research version of Claude improved the lower bound for the fraction of Riemann zeta function zeros satisfying the Riemann hypothesis from 41.6% to 67.2%, demonstrating rapid progress in AI mathematical capabilities.

An AI-generated formal proof in Lean that claimed to disprove the Collatz conjecture actually exploited two bugs in the Lean kernel, now patched. Lean creator Leo de Moura warns this will keep happening as AIs are good at finding soundness bugs.

This article presents Demonstrandum, a verification-first multi-agent AI mathematics pipeline that produces mechanically checkable artifacts, including refutations and proofs of conjectures with Lean 4 kernel verification.

Introducing Leanstral 1.5, a 119B parameter (6B active) open model for formal proof engineering in Lean 4, achieving 100% on miniF2F, state-of-the-art scores on PutnamBench and FATE benchmarks, and discovering previously unknown bugs in open-source repositories.

Mistral AI releases Leanstral 1.5, an updated Lean 4 formal proof engineering model optimized for automated theorem proving and autoformalization, with 119B total parameters and 6.5B active parameters.

Anthropic reports that an unreleased research version of Claude improved a longstanding lower bound related to the Riemann hypothesis, raising it from 41.6% to 67.2%, with a formally verified proof.