Translation. Region: Russian Federation –
Source: People's Republic of China in Russian – People's Republic of China in Russian –
An important disclaimer is at the bottom of this article.
Source: People's Republic of China – State Council News
BEIJING, Nov. 28 (Xinhua) — Chinese artificial intelligence (AI) company DeepSeek has released DeepSeekMath-V2, a groundbreaking AI mathematical reasoning model that sets new performance standards and pushes the boundaries of AI problem solving.
The new model, now open-sourced on Hugging Face and GitHub, introduces a new self-checking system designed to ensure not only correct answers but also logical and verifiable proofs.
The results demonstrated by the model are consistent with the gold medalist level of both the International Mathematical Olympiad (IMO) 2025 and the China Mathematical Olympiad (CMO) 2024.
Remarkably, this model also scored 118 out of 120 on the highly competitive 2024 William Lowell Putnam Mathematics Exam (an annual undergraduate mathematics competition in the United States and Canada), easily surpassing the best human score of 90.
The model's capabilities were further validated by the IMO-ProofBench test, where it outperformed DeepMind's DeepThink.
In the process, this system compares two large language models: one acts as a “generator” of mathematical proofs, and the other as a “reviewer” that carefully checks the reasoning.
According to the DeepSeek team, this mechanism addresses a key limitation of modern AI: a correct final answer does not guarantee a correct reasoning process.
DeepSeek said these breakthroughs establish self-verifying mathematical reasoning as a viable and promising avenue for developing more powerful and robust mathematical AI systems. -0-
Please note: This information is raw content obtained directly from the source. It represents an accurate account of the source's assertions and does not necessarily reflect the position of MIL-OSI or its clients.