Higgs Realtime Duplex

Higgs Realtime Duplex is Boson AI's 45B MoE full-duplex omni model: it listens, watches, speaks, and renders a face in roughly 160ms chunks while calling tools.

The official blog post landed around 2026-11-05 with an API waitlist: https://www.boson.ai/blog/higgs-realtime-duplex .

Highlights:

• Full duplex — no need to wait for the user to finish a full turn;
• 45B MoE aimed at realtime multimodal interaction;
• Targets voice assistants, digital humans, and live support;
• API waitlist today; not broadly open inference yet.

Why it matters

Half-duplex voice stacks usually wait for end-of-turn. Full duplex plus vision and face rendering pushes realtime digital-human UX further; tool calling also hooks into business systems.

Limits

Waitlist stage: latency, barge-in behavior, and multilingual quality need testing. The Higgs name is unrelated to Higgsfield video — this entry is Boson AI.

1. Read https://www.boson.ai/blog/higgs-realtime-duplex ;
2. Join the API waitlist and note quotas/regions;
3. Prototype short dialogs for barge-in, latency, and tool calls;
4. Check whether face rendering meets brand and compliance needs.