Meta rolled out voice calls and video chats with animated avatars to all US users of Muse, its consumer AI agent, on 5 November. Meta’s research blog says the avatar system runs “12 concurrent real-time sessions for the video generation on a single GB200”, the Nvidia chip, with about 870 milliseconds of latency.
The rollout came within about 90 minutes of two other consumer-agent launches the same afternoon, Perplexity’s teammates and Claude’s Personal Finance, by our arithmetic from the three companies’ posts on X.
Business Pill · Distribution Advantage
A short explainer of distribution advantage: why reaching users where they already are can matter as much as the product. It teaches the general idea only and does not discuss this story.
The key insight: As we read it, Meta is turning Muse into a face and a voice: one Nvidia GB200 serves 12 live avatar sessions at about 870 milliseconds of latency, by Meta’s own measurement, and the US rollout landed within about 90 minutes of Perplexity’s and Claude’s agent launches, by our arithmetic.
Muse posted on X at 18:35 UTC: “Introducing Voice. Talk to your Muse and Muse talks back”. Meta’s chief AI officer, Alexandr Wang, wrote two minutes earlier: “muse voice is out for everyone in the US!” and “you can now phone call and video chat with your muse whenever you need.”
Meta’s AI research account said it previewed Muse Realtime Avatar at Connect, and that “Today, we’re rolling it out in the @Muse app in the US, giving people a more natural way to engage with their personal agents.”
TechCrunch reported that the real-time conversations work on mobile, web and Mac and that Meta says they will later reach its AI Glasses and the planned “Muse Charm” device. It also reported that users can customise the avatar, and that Muse does not see the user but hears the conversation and reacts to it. TestingCatalog reported that Muse is also rolling out in Mexico.
Muse, Meta’s consumer AI agent account, announced Voice at 18:35 UTC on 5 November: talk to your Muse and Muse talks back. The piece uses it for the timing of the US rollout.
During Connect, we previewed Muse Realtime Avatar, our real-time embodiment technology that turns Muse Realtime Voice into expressive, interactive avatars.
Today, weโre rolling it out in the @Muse app in the US, giving people a more natural way to engage with their personal agents.
Meta’s AI research account said Muse Realtime Avatar, previewed at Connect, is rolling out in the Muse app in the US. The piece takes the avatar’s performance figures from the research blog this post links to.
What Meta Says the Avatar Costs to Run
Meta’s research blog, first published for Connect, says the avatar “streams 448×768 portrait video at 25 frames per second with approximately 870 ms of latency”, measured from the end of the user’s turn to the first byte of the reply.
It says distillation cut the work per video chunk from 120 to 2 neural function evaluations, “a 60x reduction”, and that other optimisations “increase serving capacity by 8x relative to the two-step BF16 baseline”. That is where the 12 sessions per GB200 figure comes from. Meta also says it watermarks the generated video with its Video Seal tool.
In Meta’s own rater study, its blog says, Muse Realtime Avatar was preferred over Runway Characters by 78% to 22% and over HeyGen LiveAvatar by 88% to 12%. The comparison was run by Meta, not an independent tester.
Three Agent Launches in One Afternoon
Perplexity posted its teammates launch at 17:05 UTC and Claude posted Personal Finance at 18:00 UTC; the Muse Voice post followed at 18:35 UTC. That is three consumer-agent launches in about 90 minutes, by our arithmetic. We covered Perplexity’s teammates and Claude’s Personal Finance earlier on Thursday.
As we read the rollout, Meta’s bet is presence rather than a paid tier: the avatar and voice ship inside a free consumer app.
As we read the blog, the cost work matters as much as the model: Meta says distillation and serving changes cut the work per video chunk 60x and raised serving capacity 8x.
As we read the same-day launches, the three companies are testing different things people may pay for or use daily: long-running tasks, financial answers and conversation.
The framework says: “Model capability is exponential; product capability is approximately linear.”
As we read it, voice calls and avatar video chats are product work on top of models Meta already had: the avatar was previewed at Connect in September and reached all US users on 5 November. The framework does not discuss Muse.
What Is Not Established
Meta has not said how many people use Muse Voice or what video calls cost it per user. The 12-session and 870-millisecond figures are Meta’s own measurements, and the preference results come from Meta’s own study. The Mexico rollout is TestingCatalog’s report.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Meta opened Muse voice and video calls to all US users on 5 November, saying one GB200 serves 12 live avatar sessions at about 870 milliseconds of latency. It came within about 90 minutes of Perplexity’s and Claude’s agent launches, by our arithmetic.
A note on sourcing. Product details are from Meta’s research blog, its posts and Alexandr Wang’s post on X, and TechCrunch’s report of 5 November 2026; the Mexico rollout is TestingCatalog’s report; the 90-minute figure is our arithmetic from the launch posts. The performance and preference figures are Meta’s own.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
