Xiaomi Robotics on July 14 released and open-sourced Xiaomi-Robotics-U0 (arXiv:2607.11643): a 38B-parameter multimodal autoregressive model, initialized from BAAI EMU3.5, which packs text-to-image, image editing, embodied scene generation, embodied transfer, and embodied video generation into a single next-token objective. Code, dual U0 and U0-FlashAR weights, and VisionTokenizer are simultaneously open-sourced on GitHub, Hugging Face, and ModelScope under Apache-2.0.