Gemini Deep Think at IMO 2025 (gold-medal score)
Advanced Gemini Deep Think scored 35/42 at IMO 2025, a gold-medal score, working end to end in natural language within 4.5 hours; IMO-graded.
Last year's AlphaProof needed Lean translation and days of compute. This year one Gemini model reasoned in parallel (Deep Think), trained with new RL on multi-step proof data, given curated solutions and IMO hints in its instructions.
- Date
- Monday, 21 July 2025
- Lab
- Google DeepMind
- Kind
- paper
- Access
- research preview
Figures
| Measure | Value | Measured by |
|---|---|---|
| IMO 2025 score | 35 / 42 gold-medal threshold; graded by IMO coordinators | IMO coordinators |
IMO states its review does not validate the model or process. OpenAI also claimed gold-level performance at IMO 2025 (company claim; see Wikipedia IMO page for the dispute). The submitted model differs from the one later shipped as Deep Think in the app (bronze-level on the same benchmark per Google).
Sources
- deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-m
- en.wikipedia.org/wiki/International_Mathematical_Olympiad
This record was checked against its sources on 6 October 2026. How we check