Chinese artificial intelligence company DeepSeek has announced the launch of Math-V2, a groundbreaking open-source AI model that has achieved gold-level scores at the prestigious International Mathematical Olympiad (IMO). The Hangzhou-based start-up revealed the model to the public on November 28, making it available under a permissive license that allows developers to modify and repurpose the system. Math-V2 is accessible on popular developer platforms Hugging Face and GitHub.
Since its inception in 1959, the IMO has become recognized as the most prestigious mathematics competition in the world. The event poses extraordinarily challenging problems that demand deep insight, creativity, and rigorous reasoning from both human participants and AI systems, as noted by Huang Yichen, an AI researcher at Harvard University, and Yang Lin, a computer science professor at UCLA. Only about 8 percent of participants typically earn a gold medal, underscoring the competition’s difficulty.
DeepSeek reported that its Math-V2 model has secured gold-level scores not only in this year’s IMO but also in the 2024 Chinese Mathematical Olympiad. The company asserted that this achievement will enhance global access to advanced mathematical AI tools, arriving just months after notable competitors like Google DeepMind and OpenAI reached similar milestones with their proprietary models.
“Imagine owning the brain of one of the best mathematicians in the world for free to explore it for research, fine-tune it, optimise it and run it on your own hardware,” said Clement Delangue, co-founder and CEO of Hugging Face, in a post on X. “No limitations … no company or government to take it back. That’s democratisation of AI and knowledge at its best, literally.”
Researchers at DeepSeek emphasized that Math-V2 distinguishes itself from many existing AI systems that often focus on strong performance on benchmark tests while neglecting the improvement of underlying reasoning skills. The model incorporates a “self-verification” process, allowing it to check its answers even for mathematical problems that lack known solutions. This capability addresses a significant limitation of current AI systems, which tend to excel only on tasks with easily verifiable solutions, according to the researchers in a Hugging Face blog post. They suggested that this innovative approach could lay the groundwork for more advanced mathematical AI systems and significantly impact scientific research if further developed.
In contrast to DeepSeek’s open-access approach, Google DeepMind’s gold medal-winning model is currently available only to subscribers of its premium Ultra plan. Meanwhile, OpenAI CEO Sam Altman has indicated that their experimental gold medal-winning model will not be publicly accessible for several months.
As AI continues to evolve and play an increasingly prominent role in various fields, the launch of Math-V2 could pave the way for a new era of mathematical exploration and innovation. By democratizing access to high-level mathematical capabilities, DeepSeek not only enhances the landscape of AI research but also raises the prospect of broader applications in education, engineering, and other scientific disciplines.
See also
Amazon Web Services Powers AI Growth with $132B Revenue Run Rate and 20% Surge
T1 and Faker Accept Elon Musk’s Challenge to Face AI Grok in 2026 League of Legends Match
Maximize Learning: Essential Do’s and Don’ts for Using AI in Schoolwork
Over 1,000 Amazon Employees Protest AI Rollout, Cite Democracy and Climate Concerns
Orieloris Exchange Leverages AI to Revolutionize Global Capital Allocation and Trading Efficiency
