BeingBeyond Releases Being-H0.7: A General Embodied World Model Capable of Predicting Dynamic Trajectories and Understanding Physical Laws

BeingBeyond today released its third flagship model Being-H0.7, which expands the data scale to 200,000 hours of human video and proposes a new paradigm of world model based on latent space reasoning.

BeingBeyond Technology Co., Ltd. was founded in May 2025 by Lu Zongqing, a tenured associate professor at Peking University's School of Computer Science. In the past six months, it has released Being-H0 and H0.5, the world's first embodied models pre-trained on 1,000 and 10,000 hours of human video respectively.

Being-H0.7 ranks first globally in 6 international authoritative evaluations. It is also the first general world model covering seven key dimensions: cross-ontology, cross-scenario, continuous dynamics, fluids, flexible objects, physical laws, and context reasoning. It can perform tasks like tracking and catching fast-moving objects, and precise fluid control.