Been a while since we've had a paper on provers. This "Defense-in-Depth Verifier" is actually a clever trick. Most of the paper is dedicated to defeating reward hacks. An exemplary work on what actually goes into "RL environments".
RyanLee@RyanLeeMiniMax
With the MaxProof framework, M3 exceeded the human gold-medal threshold on both sets. In this paper, we go deeper into the technical path behind our progress in mathematical proof: improving the base model, aligning a verifier, building refinement capability, and designing the test-time scaling framework MaxProof.
Here it is 《MaxProof: Scaling Mathematical Proof with Generative-Verifier RL and Evolutionary Search》 https://huggingface.co/papers/2606.13392
1:45 AM · Jun 12, 2026 · 3.6K Views
