In a landmark evaluation of AI reasoning, OpenAI's unreleased o3-preview model solved 47 out of 50 challenging Olympiad-level math problems, outperforming top open-source competitors from the AIMO Progress Prize 2. However, the combined efforts of all Kaggle participants matched this score, revealing a shrinking performance gap between proprietary and open-source AI in advanced mathematical reasoning. This experiment underscores rapid advancements in AI's problem-solving capabilities while highlighting the critical role of compute resources and model optimization.