IT之家 · 智能时代T2·Models
78 pts
Zhiyuan AI has released WITA-Omni, billed as the industry's first embodied omni-modal end-to-end interaction model, adding tactile and proprioceptive inputs plus robot action and facial-expression outputs beyond audio and video. Built on a streaming full-duplex architecture, it delivers interaction latency as low as roughly 400 ms and about 600 ms on average, nearly 3x faster than industry norms, while supporting wake-word-free omni-modal rejection and streaming simultaneous interpretation across 100+ languages and dialects.