Zhiyuan has officially released WITA-Omni (Silicon Photonics Motion Language Omnimodal Large Model), the industry's first embodied omnimodal end-to-end interactive large model. Deeply trained on the robot's proprioceptive perception data, the model accepts not only audio-visual input and output but also additional inputs such as touch and body state, along with outputs including robot actions and facial expressions. Built on a streaming full-duplex architecture, it can listen, think, respond, move, and form expressions all at once, achieving interaction response times as fast as roughly 400 milliseconds and about 600 milliseconds on average—nearly three times faster than the industry average. Its omnimodal rejection capability enables natural conversation throughout without the need for wake words, drawing on multi-dimensional information to decide whether to respond and thereby significantly reducing false triggers in interfering scenarios. The model supports seamless communication in over a hundred languages and dialects without switching, enabling streaming simultaneous interpretation and bidirectional translation. Its emotional expressiveness is remarkably humanlike, capable of conveying dozens of emotions at varying intensities and visualizing the related content in real time on a facial screen, while continuously perceiving changes in its surroundings and proactively initiating interactions at appropriate moments. A preview version of the model previously topped the DailyOmni omnimodal understanding leaderboard, with an overall score surpassing many leading models at home and abroad and a first-place ranking in six of eight sub-indicators.