DeepSeek has unveiled DeepSeekMath-V2, a next-generation mathematical reasoning mannequin constructed round a self-verifying coaching framework. Developed on high of DeepSeek V3.2-Exp-Base, the system makes use of a large-language-model–powered verification module to robotically test the correctness of generated mathematical proofs, whereas repeatedly refining its efficiency utilizing more and more difficult, high-difficulty samples.
The mannequin has demonstrated top-tier outcomes throughout main competitions—gold-medal-level efficiency on IMO 2025and CMO 2024, and a formidable 118/120 on the Putnam 2024 examination. In line with the staff, these outcomes validate the feasibility of “self-verifying reasoning pathways,” providing a promising course for constructing reliable, high-reliability mathematical intelligence programs.
Each the code and mannequin weights have been open-sourced and at the moment are out there on Hugging Face and GitHub.
Elevate your perspective with NextTech Information, the place innovation meets perception.
Uncover the newest breakthroughs, get unique updates, and join with a world community of future-focused thinkers.
Unlock tomorrow’s traits at the moment: learn extra, subscribe to our publication, and develop into a part of the NextTech neighborhood at NextTech-news.com
