DeepSeek, the Chinese tech company, has returned to the spotlight after announcing the launch of Math-V2, the first open-source AI model capable of matching the performance of gold medalists in the International Mathematical Olympiad (IMO), one of the toughest math competitions in the world.
The model is now available to researchers and developers on platforms like Hugging Face and GitHub, under a flexible license that allows full reuse and development—an announcement experts describe as a major blow to tech giants that continue to lock their advanced models behind closed systems.
Since its inception in 1959, the IMO has been known as a true test of deep intelligence, requiring not only correct answers but also creatively structured solution methods. Only about 8% of human participants achieve gold, making this level of performance a major challenge even for cutting-edge AI systems.
DeepSeek’s achievement comes months after Google DeepMind and OpenAI revealed similarly capable models, but with strict access limitations. In contrast, the Chinese model is freely available—something the CEO of Hugging Face described as “the purest form of AI democracy.”
In a blog post on Hugging Face, DeepSeek explained that its goal wasn’t just to improve performance on math benchmarks, but to strengthen the model’s self-reasoning abilities, enabling it to verify its own solutions even for problems with no known answer—a breakthrough that could push AI research into entirely new scientific territories.
While Google opted to release its top-performing model through a paid subscription, and OpenAI’s Sam Altman said their experimental model won’t be public “for several months,” DeepSeek has taken the opposite approach—offering the world an unprecedented opportunity to explore the power of mathematical AI without restrictions.
