Google DeepMind earns gold in International Math Olympiad with new Gemini AI

This year's International Math Olympiad (IMO) highlighted the brilliance of some of the world's brightest young mathematicians. Competing against advanced AI models, including Google's latest, Gemini Deep Think, the event showcased a unique blend of human intellect and artificial intelligence. Google reported that their AI model was rigorously tested under the same conditions as human contestants and managed to secure an impressive gold medal by answering five out of six questions correctly. In contrast to OpenAI's approach, Google adhered strictly to the IMO's competition rules. Last year, their AI, which combined the capabilities of AlphaProof and AlphaGeometry 2, achieved silver medal status by correctly solving four out of six problems. It's noteworthy that only a fraction of human participants earn medals at the Olympiad. For the 2025 competition, Google DeepMind collaborated with the IMO to ensure their AI models would be officially evaluated by the event's coordinators. They arrived prepared with the innovative Gemini Deep Think, which was introduced earlier this year. This model represents a significant advancement in simulated reasoning, allowing it to process multiple lines of reasoning simultaneously and integrate diverse results before arriving at a final response. Thang Luong, a senior scientist at DeepMind, emphasized that this marks a notable shift from their previous year's efforts. In 2024, human experts were needed to convert the natural language questions into a specialized format, and later, to interpret the AI's output. In contrast, Gemini Deep Think operates in natural language from start to finish, although it was not specially tailored for mathematical problem-solving.