Chinese AI firm DeepSeek has published a new research paper outlining a more efficient method for developing artificial intelligence, the company said on Tuesday.

The paper, co-authored by DeepSeek founder Liang Wenfeng, introduces a framework called manifold-constrained hyperconnection (mHC). According to the authors, the framework is designed to improve scalability while reducing the computational power and energy required to train advanced AI systems.

DeepSeek said its next-generation flagship system, R2, is expected to launch around the Lunar New Year in February.