AI from Google and OpenAI hit gold-medal level at the 2025 Maths Olympiad: 35 of 42 points
Two AI systems solved 5 of 6 problems with plain-language proofs, in the time limits. The contest graded Google's; OpenAI chose its graders.
Best AI score at the International Mathematical Olympiad
Points out of 42 on the year's IMO paper, for results graded by or reported as graded by IMO coordinators.
What this measures: The International Mathematical Olympiad is the world's top math contest for high-school students: six proof problems, 42 points in total. Gold usually needs about 29–35.
- Best AI score
- Reference level
Saturated: AI systems were reported to reach a perfect 42/42 in 2026, so the contest can't show further progress. The next frontier is open research problems. Now watch: FrontierMath Erdős and Erdős problems solved.
Gold-medal cutoffs: 35 (2025), 29 (2026). The 2026 results were reported by Chinese media and the companies; we found no statement published by the IMO itself.
Show data table and sources
| Date | Label | Value |
|---|---|---|
| 25 Jul 2024 | AlphaProof + AlphaGeometry 2 | 28 points |
| 21 Jul 2025 | Gemini Deep Think | 35 points |
| 22 Jul 2026 | dots-note-3.0 / Huawei Celia | 42 points |
| Perfect score | 42 points |
- Advanced version of Gemini with Deep Think officially achieves gold-medal standard (Google DeepMind)
- AI catches up with humans to score 100% at top maths contest (France 24 / AFP)
- dots-note-3.0 IMO 2026 perfect score (Chinese) (China Daily)
The International Mathematical Olympiad (IMO) is a contest for school students: six proof problems, written in 4.5-hour sessions. At IMO 2025, an advanced version of Google’s Gemini Deep Think fully solved problems 1–5 for 35 of 42 points, exactly the gold-medal cutoff.
It worked in plain language from the official problems, inside the time limits. IMO coordinators graded and certified the proofs with the same rules used for students. IMO President Gregor Dolinar called them “clear, precise and most of them easy to follow”.
Two days earlier, OpenAI said an unreleased experimental model had also scored 35/42 under the same time limits, with no tools or internet. Three former IMO medallists graded each proof independently. OpenAI was not an official participant.
Neither system solved Problem 6, the hardest. A year earlier, Google’s AlphaProof and AlphaGeometry 2 reached silver level (28/42), but only after people translated the problems into Lean, a language computers can check, and some problems took days of computing.
What it means
In one year, AI went from needing humans to translate olympiad problems to writing the proofs itself, under exam conditions. It matched the gold-medal line, not the best students. Neither model was available to the public at the time.
Caveats
- OpenAI’s score is self-reported and graded by people it chose. It is not certified by the IMO.
- Google says its system had access to a curated set of high-quality solutions and general hints.
- Neither lab disclosed how much computing it used.
- “Gold level” means matching the cutoff, not beating the best humans: five students scored a perfect 42.