Shengshu Technology announced its general world model strategy and launched the unified world model Motus.

Mars Finance reported on April 2nd that at the recent 2026 Zhongguancun Forum Annual Meeting, Professor Zhu Jun, founder of Shengshu Technology and Vice Dean of the Institute for Artificial Intelligence at Tsinghua University, systematically elaborated on the company's strategic layout in the field of general world models. This strategy uses a base world model as its core foundation, extending upwards to a dual-track system connecting digital and physical spaces. In the digital space, Vidu, a video large-scale model product, is built based on the world generation model; in the physical space, Motus, a unified world model product, is built based on the world action model. Motus, based on the UniDiffuser unified modeling framework, integrates visual language knowledge, video dynamics knowledge, and motion skill knowledge. Experimental data shows that compared to the internationally leading VLA model, Motus achieves an absolute improvement of 35.1% in the average success rate across 50 tasks, a 13.55-fold increase in data efficiency, and demonstrates stronger multi-task generalization capabilities. In complex real-device tasks such as CAPTCHA operations, chess decision-making, and flexible object grasping, the model exhibits near-human-level decision-making logic and execution stability. (Wide-angle observation)