Chinese artificial intelligence company DeepSeek has released the world’s first open AI model to test and score a gold medal-level performance at the annual International Mathematical Olympiad (IMO), the Hangzhou-based start-up said on Thursday.
DeepSeek has made widely available its Math-V2 model, which was open-sourced on developer platforms Hugging Face and GitHub, under a permissive licence that allows users to repurpose the model.
Held annually since 1959, the IMO is widely regarded as the world’s most prestigious maths competition, including for AI systems, because of the complicated nature of the questions involved, which require “deep insight, creativity and rigour”, according to Harvard University AI researcher Huang Yichen and Yang Lin, a computer science professor at the University of California, Los Angeles.
To achieve top scores, AI systems must demonstrate the reasoning behind their outputs, rather than just provide simple answers. Around 8 per cent of human IMO participants achieve a gold medal on the test.
DeepSeek’s accomplishment was expected to lower the barrier for developers around the world to access a powerful maths AI model, a few months after leading US AI giants Google DeepMind and OpenAI achieved the same feat at IMO with their proprietary models.
The Chinese firm claimed its model achieved gold-level scores on questions in both this year’s IMO and the 2024 Chinese Mathematical Olympiad.
