Now it seems the best way to tell if a frontier model is benchmaxxed is to check if it can autonomously solve a major open mathematical problem.