This market resolves YES if an AI model that was published (made publicly available or described in detail in a public paper) before the start of the 2025 International Mathematical Olympiad (IMO) achieves a score that would qualify for a gold medal at the 2025 IMO.
To qualify for a gold medal at the IMO, a contestant typically needs to score in approximately the top 1/12 of all participants. In recent years, this has corresponded to solving 4-5 out of 6 problems correctly.
The AI must solve the actual 2025 IMO problems within the standard time constraints (4.5 hours per day over 2 days) or demonstrate equivalent performance on these problems after they are released.
Resolution will be based on official announcements from the AI's developers, verification from reputable sources, and/or relevant academic papers.
Some claim of the feat must be made by September 1st. The market will extend if the claim is made but independent verification is in flight.
Sources:
People are also trading
How much scaffolding is allowed? This work claims to have solved it with Gemini 2.5 Pro + scaffolding: https://github.com/lyang36/IMO25/blob/main/IMO25.pdf
I don't think this work counts since it they mention they gave (small) hints for Problems 1 and 2, and they also don't seem to mention time limits. However, this still raises the question of whether the sort of scaffolding they do (which was not published before the 2025 IMO) would have disqualified this model system for this market or not.
I contend that this amount of scaffolding should disqualify the work (unless the scaffolding was announced before the 2025 IMO) for the purposes of this market. But I am not a disinterested party.