Dexmal Launches Embodied MaaS and DexOS, Solving Model Scaling Into Real-World Scenarios

Dexmal, formerly known as Yuanli Lingji, unveiled a comprehensive suite of embodied AI products at its inaugural developer conference. The centerpiece is the DM0.5 model, which doubles parameters to 4 billion and is trained on 150,000 hours of data, including 50,000 hours of real robot data. Three architectural innovations distinguish DM0.5, including native memory support for up to 60 seconds of context and a two-tier reasoning system that separates task planning from motion generation. DexOS, a standardized operating system for robotics, facilitates model integration across multiple robot platforms.