Moxin Technology has launched MoWorld 2.0, a 4D world model that simulates physics and was trained and deployed natively on Huawei Ascend hardware. The company unveiled the model during a World Internet Conference forum in Wuzhen on November 4, according to IT Home. This release follows the company's launch of MoWorld 1.0 in July.
Moxin Technology states that the primary advancement in version 2.0 is the integration of physics. The model incorporates gravity, elasticity, buoyancy, lighting, and heat, as well as the properties of fluids, soft bodies, rigid bodies, and deformable materials into how it builds and interacts with spaces. The model produces 4D Gaussian assets that can be reused repeatedly and rendered in real time.
Because the model accounts for these physical factors, Moxin Technology says it generates more than just the appearance of a space. It claims the model reasons in real time regarding how external forces act on objects, how those objects interact, and how the environment changes. One demonstration showed a breeze turning the pages of a book and a bookmark falling to the floor, which Moxin says required the model to account for the weight of the bookmark, the rigidity of the floor, the flexibility of the paper, and the direction and strength of the wind. Other demonstrations showed scenes where time could be sped up or slowed down and where users could interact with and edit the environment.
The model was trained and deployed on a fully domestic Ascend computing base. Moxin Technology says it collaborated with Huawei teams on multi-NPU scheduling, inference system optimization, and long-sequence training. Based on the company's own tests and estimates, Moxin Technology says the inference cost can drop to as little as 30% of a GPU setup of the same scale.
The company also reported speed improvements. Video inference increased from a peak of about 50 frames per second in version 1.0 to about 70 frames per second. Rendering of 4D and 3D Gaussian assets exceeds 500 frames per second, while tests on phones reached about 100 frames per second. Moxin Technology previously released a lightweight MoWorld model for Huawei's Mate 90 phone.
Moxin Technology says it is currently working with partners on validation and deployment across several sectors. For embodied AI, the model builds virtual environments with spatial structures and basic physical feedback to allow robots to train and retry actions in simulation. In autonomous driving, it generates long-tail scenes including potholes, sudden obstacles, snow, and rain. Other applications include urban planning, virtual tourism, industrial digital twins for mines and factory lines, and reusable assets for film and game production that support weather changes, object editing, and free camera movement.
