Zhiyuan's full-modal model tops DailyOmni rankings

Zhiyuan's self-developed native multimodal model WITA-Omni Preview topped the DailyOmni rankings with a score of 85.21. It achieved first place in six out of eight sub-indicators, beating models like Qwen, Gemini, Doubao, and Nvidia. DailyOmni tests audio-video timing alignment and cross-modal reasoning using real-world materials. WITA-Omni uses a Thinker-Talker-Actor architecture, elevating physical actions and expressions to primary outputs alongside speech. The model was trained on millions of hours of open-source and proprietary multimodal data, with a focus on real-world interaction scenarios.