DeepSeek opens access to gold-level maths AI
Open release of Math-V2 promises broader access to advanced reasoning systems, following its gold-level results in top international mathematics competitions.
Chinese AI firm DeepSeek has released the first open AI model capable of achieving gold-medal results at the International Mathematical Olympiad. Math-V2 is now freely available on Hugging Face and GitHub, allowing developers to repurpose it and run it locally.
Gold-level performance at the IMO is remarkably rare, with only a small share of human participants reaching the top tier. DeepSeek aims to make such advanced mathematical capabilities accessible to researchers and developers who previously lacked access to comparable systems.
The company said its model achieved gold-level scores in both this year’s Olympiad and the Chinese Mathematical Olympiad. The results relied on strong theorem-proving skills and a new ‘self-verification’ method for reasoning without known solutions.
Observers said the open release could lower barriers to advanced maths AI, while US firms keep their Olympiad-level systems restricted. Supporters of open-source development welcomed the move as a significant step toward democratising advanced scientific tools.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
