Overview
- Moxin officially unveiled MoWorld 2.0 at a World Internet Conference forum as an upgraded 4D world model that embeds gravity, elasticity, buoyancy, light, heat and rigid/soft/fluid object properties into scene and object representations.
- The model performs real-time physical reasoning for object interactions, with demonstrations showing fine details such as wind turning pages and a bookmark falling while accounting for material flexibility and contact forces.
- Moxin reported that MoWorld 2.0 has been trained and deployed on Huawei’s Ascend computing platform and that the company is working with Huawei teams on long-sequence training, multi‑NPU scheduling, and inference optimizations.
- Company benchmark results claim faster rendering and lower inference cost—video inference up to about 70 fps, scene rendering above 500 fps, phone rendering near 100 fps, and inference cost reduced to around 30% of similar GPU solutions—but those figures come from internal tests and lack independent verification.
- Moxin is running phased pilots with partners across robotics, autonomous driving, industrial digital twins and entertainment/tourism to validate scene editing, reusable 4D assets, and potential workflow and cost changes as domestic AI hardware is applied at scale.