LHO

The Ledger · Entry 011 · Mathematics

Olympiad mathematics (IMO)

Ceded to machinesSince 2025Criterion performanceLast reviewed 2026-08-20

Event

An advanced Gemini Deep Think model scores 35 of 42 at IMO 2025, gold-medal level, graded by IMO coordinators; OpenAI reports the same score for an experimental model, graded by former medalists.

The defined task is the six IMO problems under contest time (two sessions of four and a half hours), in natural language, with proofs read by human graders. Both systems solved problems 1 to 5 and neither solved problem 6; five human contestants scored a perfect 42 that year, so the best humans were still ahead on the paper itself. That last gap closed in July 2026 at the Shanghai olympiad, where Huawei's Celia and Xiaohongshu's dots-note-3.0 each scored 42 of 42 under the official grading, having received the problems only after the contestants had finished and worked to a time limit with human intervention forbidden. Seven of the 666 contestants matched them. Four other perfect scores reported that week came from a private test graded by machine and belong in a different column. Gold-medal level is ceded; the frontier of research mathematics, where problems are not chosen to be solvable in hours, is a separate question.