Gemini by Google Deepmind wins gold at the International Mathematics Olympiad 2025

The 66th International Mathematics Olympiad (OIM) in Australia was very important for artificial intelligence. Therefore, For the first time as part of the competition. Moreover, the Gemini model “of Google Deepmind obtained the gold medal by solving five of the six exceptionally difficult problems of Olympiad and obtaining 35 points out of 42, which corresponds to the gold threshold fixed by the omi rating committee.

This is a clear improvement compared to last year. In addition, where AlphaProof and Alphageometry 2 of Deepmind had reached the level of the silver medal by solving four problems and obtaining 28 points. For example, Previous systems required a translation between natural language and formal evidence languages, and the calculation of solutions took several days.

This year. Gemini worked from start to finish using only natural language, producing complete and rigorous solutions directly gemini google deepmind wins gold from the OMI official descriptions in the 4.5 -hour competition window. OMI official correctors. who also evaluate human competitors, qualified Gemini’s responses as “clear” and “precise” Deepmind has attributed this progress to advanced strengthening learning techniques, a database of mathematical solutions and new parallel reasoning methods that allow you to simultaneously explore several ways of solution. While other AI systems. including those of the Openaiont obtained from similar unofficial results, Gemini’s evaluation was officially certified by Olympiad coordinators, which is a world first for autonomous mathematical systems.

Deepmind plans to put the Deep Think model available to certain mathematicians before deploying it more widely. while continuing to promote natural language and formal reasoning agents for mathematics research.