← Retour au blog
tech 12 June 2026

MaxProof: Maximizing Mathematical Proof with AI

MaxProof redefines mathematical proof using generative AI and RL. This system has outperformed human scores in IMO and USAMO competitions with an innovative model.

Article inspired by the original source
Maxproof ↗ arxiv.org

Introduction

AI continues to push the boundaries of innovation across various fields, and mathematics is no exception. MaxProof, a recent breakthrough, uses AI to solve complex mathematical proofs, even surpassing human performance in some cases. But how exactly does MaxProof work, and what does it mean for the future of mathematics?

What is MaxProof?

MaxProof is a revolutionary approach to solving mathematical proofs, developed as part of the MiniMax-M3 series. The system relies on three main capabilities: proof generation, proof verification, and critique-conditioned proof repair.

Proof Generation and Verification

MaxProof's proof generation aspect uses reinforcement learning (RL) techniques to create possible solutions to mathematical problems. Once generated, these proofs undergo a rigorous verification process designed to minimize false positives. This ensures that only valid solutions are retained.

Proof Repair

Beyond mere generation and verification, MaxProof incorporates a proof repair function. By leveraging critique-conditioned methods, the system can refine proofs that initially fail the verification process, thus continually improving its performance.

Impressive Performance

During testing, MaxProof achieved impressive results: 35/42 on the IMO 2025 and 36/42 on the USAMO 2026, surpassing the human gold medal threshold. These results not only demonstrate AI's capability to solve complex problems but also its potential to transform how we approach mathematics.

Implications for Education and Research

MaxProof opens new avenues for education and research in mathematics. By automating proof solving, researchers can focus on exploring new theorems and concepts, while students benefit from a powerful tool to understand complex proofs.

Conclusion

MaxProof is not just a step forward for AI and mathematics; it's a revolution. By combining the power of machine learning and reinforcement, it redefines what's possible in the realm of mathematical proofs. Let's discuss your project in 15 minutes.

MaxProof IA mathématiques preuve apprentissage par renforcement
Deepthix newsletter · 100% AI · every Monday 8am

An AI agent reads tech for you.

Our AI agent scans ~200 sources per week and ships the best articles to your inbox Monday 8am. Free. One click to unsubscribe.

Visit the newsletter page →

Want to automate your operations?

Let's talk about your project in 15 minutes.

Book a call