Google Deepmind announced yesterday that an advanced version of its artificial intelligence system (AI) Gemini Deep Think has reached a gold medal level performance at the International Mathematics Olympiad (IMO) that was held this 2025 in Australia.
The AI system perfectly solved five of six problems, obtaining 35 of 42 possible points, reaching the gold medal threshold. This represents a significant improvement regarding the performance of the silver medal last year, when the combined Alphaproof and Alphageometry 2 Google systems obtained only 28 points.
The president of the IMO, Gregor Dolinar, confirmed the achievement stating: “Google Deepmind has reached the desired milestone, obtaining 35 of 42 possible points – a gold medal score – their solutions were amazing in many aspects. IMO evaluators found them clear, precise and most easy to follow.”
Unlike previous attempts that required experts to translate problems into specialized languages, this version of Gemini operated completely in natural language, producing mathematical evidence directly from the official approaches of the problems within the time limit of four and a half hours of the competition.
The advance used an advanced version of Gemini Deep Think with improved reasoning capabilities, including parallel thought
which allows the model to explore multiple paths of solution simultaneously. The system was trained for months using novel learning techniques for reinforcement and was given access to high quality mathematical problems.
On Saturday, the American Openai also announced that her deep learning model had finally won a gold medal in the most important pre -university math competition in the world.