Maxproof
4 hours ago
- #test-time scaling
- #machine learning
- #mathematical proof
- MaxProof is a framework that scales competition-level mathematical proof using population-level test-time scaling.
- The system is built on the M3 model, which integrates three capabilities: proof generation, proof verification, and critique-conditioned proof repair.
- A defense-in-depth generative verifier is engineered to minimize false positives during training.
- At test time, MaxProof uses the model as a generator, verifier, refiner, and ranker to search a population of candidate proofs.
- It selects a final proof through tournament selection.
- With MaxProof scaling, the M3 model achieves high scores on IMO 2025 (35/42) and USAMO 2026 (36/42), surpassing the human gold-medal threshold.