Second Progress Prize closed

The second AIMO Progress Prize has concluded and the top team on the final Kaggle leaderboard is NemoSkills, with a score of 34/50.

It was a closely contested six-month competition with innovative open-source models tackling a set of National Olympiad-level math problems specially designed for this challenge. The impressive progress demonstrated by the competitors, since the first AIMO Progress Prize, highlights remarkable advancements in large language models (LLMs), the growing number of teams, and the continuing industriousness of the Kaggle community.

The top five teams will publish their models to ensure scientific reproducibility and the proliferation of open-source AI models for mathematical reasoning. The models will be assessed by AIMO judges to ensure they meet the competition’s open-source criteria, and results will be announced shortly. The top five teams will earn a share of the prize pot, with the winning team receiving $262,144.

Third Progress Prize later this year

Following the successful completion of the competition’s second stage, the third AIMO Progress Prize will launch in Autumn 2025. The difficulty level will increase again, with the problems centred on IMO-level problems. Competition rules will be fine-tuned based on community feedback and continuing format improvements are planned.

Full details on timings, prize pot size and the improved competition format will be announced closer to launch date.

Additional evaluations planned

The AIMO’s stated objective is the development of open-source AI models capable of mathematical reasoning, with only locally run LLMs permitted to compete in the competition. To assess the gap between the performance of closed LLMs, which lead the frontier of reasoning capabilities, and open-source LLMs, we will be performing an evaluation of the second Progress Prize problems on selected unreleased closed models.

Stay tuned for results in the coming days!