RMA: an Agentic System for Research-Level Mathematical Problems

In three linesRMA is a multi-agent system for solving research-level mathematical problems. It decomposes proof solving into specialized modules (problem analysis, literature search, verification) coordinated by initializer, proposer, and verifier agents. On the First Proof benchmark (10 problems), RMA solves 8/10 and outperforms GPT-5.2R and Aletheia.

Summary generated by Claude — human-verified