The second Artificial Intelligence Mathematical Olympiad (AIMO2) competition has just concluded. https://www.kaggle.com/competitions/ai-mathematical-olympiad-progress-prize-2/overview . (I am on the advisory committee for this competition.)

The 50 test problems were designed to be more resistant to brute force methods than the AIMO1 problems, but still of the same "numerical response" form. Contestants were required to use open-source models, and were assigned a fixed amount of compute.

Final certification has not yet occurred, but the presumptive winner is NemoSkills, who solved 34/50 of the problems (for AIMO1, the best performer achieved 29/50).

We are currently evaluating performance against AIMO2 for models that are more advanced, or are allowed more compute. It appears that there are significant improvements in such cases; stay tuned for announcements.