Unifies text, images, audio, and video for Chinese-English multimodal understanding and reasoning, with fast instruct and explicit thinking modes. The paper reports leading open-source results on several omni-modal video benchmarks.