Maxproof
MaxProofgenerative-verifier RLpopulation-level test-time scalingmathematical proofMiniMax-M3 series
Author: ilreb
Date: 6/12/2026
Article Summary:
This paper presents MaxProof, a population-level test-time scaling framework for competition-level mathematical proof in the MiniMax-M3 series, using a defense-in-depth generative verifier and generative-verifier reinforcement learning.