Бенчмарк · math
IMO-AnswerBench
400 IMO-level olympiad problems graded by final-answer matching (reported avg@8); a headline math row next to AIME/HMMT. Not the IMO competition itself — a gold-medal claim is not a score here.
| Дата | Модель | Результат | Источник |
|---|---|---|---|
| 2026-07-28 | Claude Opus 5 | 100.0% | Anthropic выпустила Claude Opus 5 как экономичную альтернативу Fable |