DeepSeek makes a strong comeback with the open-source IMO gold medal-winning mathematical model DeepSeek-Math-V2.

DeepSeek has open-sourced its next-generation mathematical model, DeepSeek-Math-V2, achieving IMO gold medal-level inference capabilities. The model innovatively introduces a self-verification mechanism, employing a triple architecture of "generator-verifier-meta-verification" to enable the AI to examine and correct its own reasoning process. It achieved a high score of 83.3% in the IMO 2025 test and nearly a perfect score in the Putnam 2024 test.