WITA-Omni

WITA-Omni is AGIBOT's first full-modal, end-to-end interaction model for humanoid robots, released as WITA-Omni 1.0 at AGIBOT Innovation Day 2026 on October 29, 2026. It lets a robot hold natural spoken conversations with people while taking in what it sees and hears.

Source: @RobotDollLab on X · Event lineup: AGIBOT Hong Kong

Key points

1. End-to-end and full-modal: one model handles listening, understanding and speaking (plus visual context), instead of chaining separate speech recognition, LLM and text-to-speech steps.
2. Fast replies: response latency as low as 400 ms, about 600 ms on average, close to human turn-taking.
3. Languages: 100+ languages and 29 dialects.
4. Real-world robustness: works in noisy environments and without a wake word, which matters on factory floors, shops and reception desks.
5. Part of a stack: AGIBOT positions it as the human-robot interaction foundation model next to GO-3 Preview (embodied foundation model) and Link-U OS (robot operating system).

Where it fits

• Reception and retail humanoids that greet, answer questions and guide visitors in many languages.
• Factory and warehouse robots that need to take spoken instructions over background noise.
• Developers on AGIBOT hardware who want natural conversation without building their own speech pipeline.

Next steps

Watch AGIBOT's official channels for SDK or API availability and which robot models ship with WITA-Omni 1.0.