Current AI models struggle to solve research-level math problems, with the most advanced AI systems we have today solving just 2% of the hundreds of challenges faced.