OpenAI's o3 model currently has an 88% Overall accuracy on the matharena.ai benchmark, according to its performance on the AIME 2025, HMMT February 2025, and BRUMO 2025. This market resolves to 'Yes' if any AI model is shown to surpass this score by July 1, 2025.

Resolution criteria

This market resolves to "Yes" if any AI model surpasses OpenAI's o3 model's overall score on MathArena.ai by July 1, 2025. The overall score is determined by the model's performance across several tests of matharena.ai's choosing. Verification will be based on the official leaderboard available at MathArena.ai.

Background

OpenAI's o3 model, released in early 2025, has demonstrated exceptional performance in mathematical reasoning tasks. Notably, o3 achieved a 96.7% accuracy rate on the 2024 American Invitational Mathematics Examination (AIME) (aitide.news). Additionally, o3 set a new record on the EpochAI Frontier Math benchmark with a score of 25.2%, significantly outperforming previous models that did not exceed 2% (aitide.news). These achievements highlight o3's advanced capabilities in complex mathematical problem-solving.

Resolution criteria

This market resolves to "Yes" if any AI model surpasses OpenAI's o3 model's Overall score on matharena.ai by July 1, 2025. The Overall score is determined by the model's performance across various mathematical exams hosted on matharena.ai. Verification will be based on the official leaderboard available at https://matharena.ai/leaderboard.