DeepSeek AI Nails Several Math Olympiads

The DeepSeekMath-V2 artificial intelligence has demonstrated a mathematical ability on par with gold medalists at the 2025 International Mathematical Olympiad (IMO) and the 2024 Chinese National Mathematics Olympiad (CMO).

The key feature of this model isn't just about finding the correct answer. Its creators set their sights on the capacity for rigorous mathematical reasoning, making the deep development of logical chains and theorem proofs a key priority. This approach earned the system a near-perfect score of 118 out of 120 on the notoriously difficult 2024 Putnam competition.

The algorithm's operation is based on a closed-loop system where two modules interact. The first one proposes a solution, and the second one meticulously checks every step. If the reasoning contains gaps, the system gets a signal about it and goes back to searching for a more correct path, mirroring the work of a scientist who constantly double-checks their conclusions.

Technically, the model contains hundreds of billions of parameters and is available to the research community. Its source code and architecture are hosted on GitHub, allowing mathematicians and AI specialists to study this approach in more depth. The development of such systems capable of verifiable reasoning could, in time, help solve fundamental scientific problems where flawless logic is required.

However, DeepSeekMath-V2's success in solving abstract problems brings us back to more ground-level questions previously raised by the company's founder. Chen Deli stated that the rapid development of AI, even in complex fields like mathematics, will create significant pressure on the labor market in the coming years. Thus, every AI victory is not only a scientific breakthrough but also another argument suggesting that industry and society need to find a balance between adopting innovation and mitigating its social consequences.