Agents
Meta adds real-time voice and expressive avatars to its Muse agent, rolling out first in the US on mobile, web and Mac
Meta began rolling out real-time voice and expressive avatars in Muse on Nov. 5, US only. Figures are Meta's own; privacy details are missing.
On November 5, 2026, Meta began rolling out real-time voice conversations and expressive avatars in its Muse app, starting in the United States, on mobile, web and Mac. The avatar is driven by Muse Realtime Voice and Muse Realtime Avatar, technologies Meta previewed at Connect. Voice for Meta's AI glasses and the upcoming Muse Charm device is described as coming later, with no dates. The technical and quality figures below are Meta's own, and none has been independently verified.
The essentials
- What launched: voice conversations with Muse and an avatar that animates while it speaks, in the Muse app on mobile, web and Mac. In the app, a phone icon starts a voice session.
- Where: the United States only for now. Meta gave no date for other countries.
- What is reported: about 870 ms from the end of your turn to the first byte of the response, and portrait video at 448x768 and 25 frames per second. These are Meta's figures, with no published method.
- What is missing: pricing or usage limits for this feature, language coverage beyond "hundreds of voices", and privacy or retention rules for voice and avatar sessions.
- Later: voice on AI glasses and Muse Charm, with no dates.
What exactly did Meta release?
Two things in one: a voice mode where Muse answers aloud and can be interrupted, and an avatar that reacts while it speaks. It can appear beside the conversation, fill the screen, or stay hidden in voice-only mode. According to Meta's research post as relayed by one tracker, Muse Realtime Voice produces a stream of speech tokens that carry both the words and how they are said. Muse Realtime Avatar, described as an audio-driven diffusion transformer, reads that stream to generate video in short chunks, so lips and expressions follow the voice.
What do Meta's numbers say, and how solid are they?
Meta reports about 870 ms of latency, 448x768 portrait video at 25 fps, an invisible watermark from Meta Video Seal on generated video, and up to 12 live sessions on a single GB200. In Meta's own preference study, raters preferred Muse Realtime Avatar 78% to 22% over Runway Characters and 88% to 12% over HeyGen LiveAvatar. The raters were Meta's, and the number of raters, the prompts and the statistics were not given. The latency page also gives no hardware, network conditions or percentiles, so treat the figures as claims.
Can Muse see me during a call?
According to a TechCrunch report relayed by one outlet, Muse cannot see the user during a call and only hears them. We did not read that report ourselves. Meta's glasses plans are separate: voice on AI glasses and Muse Charm is described only as coming later.
What is not established?
- Whether the feature is free or paid, and any usage limits.
- How long voice or avatar sessions are retained, and who can access them.
- Availability outside the US, including Canada and Europe.
- Language support beyond the "hundreds of voices" claim.
- Independent measurements of latency, quality, watermark robustness or interruption handling.
- Dates for AI glasses and Muse Charm.
Summary table
| Point | What is stated as of November 6, 2026 | Source |
|---|---|---|
| Launch | November 5, 2026, Muse app on mobile, web, Mac | iPhone in Canada, ShiftDelete |
| Availability | US first, no other dates | iPhone in Canada, ai-tldr |
| Technology | Muse Realtime Voice and Muse Realtime Avatar | Meta, via trackers |
| Latency | About 870 ms, Meta-reported, no method | Meta research post via ai-tldr |
| Quality study | Preferred 78% and 88% against two rivals, by Meta's raters | Meta research post via ai-tldr |
| Glasses, Muse Charm | "Later" or "coming soon", no dates | iPhone in Canada, ai-tldr |
Frequently asked questions
Is Muse voice available outside the US?
Not yet. The rollout is US first and Meta has given no date for other countries.
Is the 870 ms latency independently verified?
No. It is Meta's figure and the published description gives no hardware, network conditions or percentiles.
Does the avatar come from a person's face?
The sources we read describe it as an avatar for Muse, not a copy of the user. Details beyond that are not stated.
Method and limits
Sources as of November 6, 2026. We read secondary coverage and trackers that relay Meta's announcement and research post. One page we tried to open (MLQ) was unreachable, so we relied on trackers and coverage for the voice details. We did not read Meta's own posts directly and tested nothing.
Sources
- iPhone in Canada: Meta's Muse gets real-time voice and expressive avatars, not in Canada yet (verified 2026-11-06)
- ShiftDelete: Meta Muse Voice lets you talk to your AI agent live (verified 2026-11-06)
- ai-tldr: Muse voice and avatars, Meta's AI agent now talks back with a face in the US (verified 2026-11-06)
- Runtimewire: Meta gives Muse a live avatar, with subsecond video responses (verified 2026-11-06)
- OrcaRouter: Muse Realtime Avatar release (verified 2026-11-06)
Spotted an error in this article? Report it · Our corrections policy
