Google and OpenAI’s AI Models Reach Gold Medal Level in International Mathematical Olympiads!

Metaverse Planet July 23, 2025Last Updated: July 23, 2025

0 2 minutes read

Google DeepMind has announced that an advanced version of Gemini, featuring superior reasoning and deep thinking capabilities, has reached the level to win a gold medal in the Mathematical Olympiads. OpenAI has also stated that one of its models has achieved this same impressive level.

Artificial intelligence technologies continue to surprise us with their daily advancements. Among the leading companies in this field, Google and OpenAI undoubtedly stand out. Now, both companies have made announcements demonstrating the remarkable heights their AI models can reach.

According to a post shared by Google’s AI department, DeepMind, an advanced version of Gemini Deep Think, with its deep thinking capabilities, has successfully demonstrated gold medal-level performance in the 2025 Mathematical Olympiads. Meanwhile, a similar achievement has been announced by another AI giant, OpenAI.

Contents show

Gemini Achieves Gold Medal Level with 35 out of 42 Points

The advanced Google Deep Think AI system, possessing superior reasoning and deep thinking skills, managed to score 35 out of a total of 42 points in the Mathematical Olympiads. This means it solved 5.83 out of 6 problems, indicating it has reached the gold medal level.

This is a clear indicator of how much AI systems have advanced in just one year. For instance, last year, Google’s AlphaProof and AlphaGeometry 2 systems had reached a level of 28 points. Google officials described Gemini’s performance as “awe-inspiring” and noted that its solutions were easy to understand, clear, and precise.

Previously, such problems had to be translated into specialized languages for AI models. However, with this version of Gemini, this was not the case. The model operated in natural language, directly solving problems based on their definitions within 4.5 hours. It also utilized its “parallel thinking” capability, allowing it to consider different solution paths simultaneously. Google plans to begin testing this version of Gemini soon with a limited number of users, primarily mathematicians. Afterward, it is planned to be offered to AI Ultra subscribers.

OpenAI Model Also Reaches Gold Medal Level

Meanwhile, ChatGPT developer OpenAI made a similar announcement. The company stated that in independently conducted tests, an unnamed “experimental model” successfully answered 5 out of 6 questions, reaching a gold medal level. While Google’s model officially entered the competition and was evaluated by the committee, OpenAI’s model was assessed by 3 independent individuals who had previously won Olympiad medals.