AGIBOT just announced WITA-Omni 1.0: an omni-modal interaction model

• Duplex conversation: listening and speaking at the same time
• Runs on-device, with end-to-end interaction latency of ~400ms
• Fuses vision, speech, real-time dialogue, facial expressions and physical actions into one continuous system
• Currently ranks at the top of several interaction benchmarks

There were a series of other announcements at AGIBOT Innovation Day today:

• Worken (Work + Token): a proposed metric for productive robot work (active robots x autonomous hours x output efficiency), shifting focus from robots built to useful work delivered.
• AHA3: 3rd-gen humanoid platform with liquid-cooled IP67 joints and OmniHand 3 dexterous hands.
• GO-3 Preview: new foundation model (World Reasoning Action Model), #1 on RoboDojo, RoboTwin, RoboColiseum.
• G2 Max: industrial robot, 40kg dual-arm lift.
• Link-U OS 1.0: open-source by end of 2026.

Impressive demo of talking over the robot to test its ability to listen while speaking.: