DeepSeek's AI model checks itself and cracks Olympiad-level math problems
What's the story
Chinese artificial intelligence (AI) start-up DeepSeek has unveiled a new open-weight AI model, DeepSeek-Math-V2. The advanced system is designed to generate and self-verify mathematical theorems using advanced reasoning skills. The company claims these capabilities were specifically developed for this task. The model weights are publicly available under the Apache 2.0 open-source license on platforms like Hugging Face and GitHub.
Model features
DeepSeek-Math-V2's architecture and capabilities
DeepSeek-Math-V2 is built on top of DeepSeek-V3.2-Exp, an experimental AI model launched by the Huangzhou-based start-up in September. The new model focuses on self-verifiable mathematical reasoning and consists of two main components: a verifier that checks mathematical proofs step-by-step, and a theorem generator capable of correcting its own mistakes. This unique design enables it to solve open problems without known solutions through self-verification as a means to scale test-time compute and perform deeper reasoning.
Competition results
DeepSeek-Math-V2's performance in mathematical competitions
DeepSeek-Math-V2 has been tested on math problems from the International Mathematical Olympiad (IMO) 2025 and CREST Mathematics Olympiad (CMO) 2024, achieving gold medal-worthy scores. It also scored an impressive 118 out of 120 on math problems from the Putnam 2024 mathematical competition. These results suggest that self-verifiable mathematical reasoning is a viable research direction for developing more capable mathematical AI systems.
Historic feat
DeepSeek-Math-V2's achievement in IMO 2023
DeepSeek's gold medal-level score on IMO 2025 problems puts it in the elite company of OpenAI and Google DeepMind, whose unreleased models posted similar results at the prestigious math competition. However, this was the first year IMO organizers formally admitted AI models in the competition, and while Google was part of that inaugural cohort, OpenAI and DeepSeek were not included.
Research implications
DeepSeek-Math-V2's potential impact on research
The recent progress made by reasoning models in the mathematical domain could help researchers solve long-standing problems in areas like cryptography and space exploration. DeepSeek's Math-V2 model is also a major milestone for the open ecosystem, which has been dominated by China. A study by MIT and open-source AI start-up Hugging Face found that downloads of new Chinese-made open models increased to 17% over the past year.