4% that AI solves the hardest IMO problem by 2025
“I'd put 4% on "For the 2022, 2023, 2024, or 2025 IMO an AI built before the IMO is able to solve the single hardest problem"”
How we judge it
An AI solves the designated hardest problem (usually #6) at a 2022-2025 IMO.
Resolution criteria are our interpretation of the claim, written before grading. Methodology
- Said
- Resolve by
- Verdict
- Correct ()
- Stated probability
- 4%
- Specificity
- 3 of 3 (precise and measurable)
- Score
- Brier 0.002 → 1.00 points
Evidence & grader’s note
- IMO 2025 official problems (Problem 6: tiling a grid)
- IMO 2025 Shortlist: Problem 6 is combinatorics problem C8
- IMO 2025 official individual results (per-problem scores)
- Google DeepMind: Gemini Deep Think IMO 2025 solutions (Problems 1-5)
- Alexander Wei (OpenAI) on X: model solved P1-P5, no solution for P6 (Jul 19, 2025)
- LessWrong: 2025 IMO gold; problem 6 discussion
- arXiv (Dec 2025): human-guided workflow with later models reaches a P6 proof after the contest
Christiano's rule picks problem 6 unless it is geometry, or problem 3 is combinatorics and problem 6 is algebra. In 2025 problem 6 was combinatorics (shortlist C8) and problem 3 number theory, so problem 6 is the designated hardest problem. It was also the hardest for contestants: 6 of 630 scored full marks (our count from the official results). Neither gold-level AI run solved it: Google DeepMind's published solutions cover problems 1-5, and OpenAI's researcher said its model produced no solution for problem 6. A December 2025 paper reported a human-guided workflow using later models that reached a proof after the contest; that does not meet the bet's conditions. Brier = 0.0016.
