Realtime AI News
Meta opens Muse video calls to all US users, giving its AI agent a face
Meta made Muse Voice available to everyone in the US, letting users call and video chat with its consumer AI agent through the Muse Realtime Avatar and Muse Realtime Voice models. The feature debuted at Connect in September, works on mobile, web and Mac, and is headed for AI Glasses and the Muse Charm wearable due before year-end.

Meta announced on November 5 that Muse Voice is now available to everyone in the United States, letting users place calls and start video chats with Muse, the company's consumer AI agent. Chief AI officer Alexandr Wang announced the rollout in a blog post and on X, sharing a demo of his own Muse in action.
The capability first appeared at Meta's Connect developer conference in September and runs on two new models, Muse Realtime Avatar and Muse Realtime Voice, which power the animated character that talks with the user. TechCrunch reported that Meta confirmed the broad US rollout on Thursday.
Muse appears as an animated character nicknamed Jolly, and users can customize it to look more like a person, an animal or some other figure. As the avatar speaks, Meta says it uses subtle expressions and gestures matched to the tone of the conversation — a clear bet that anthropomorphizing the agent will make people feel more connected to it.
In the demo, Wang had his Muse appear as a fluffy white dog; when he joked that his meme posting on X might have annoyed his boss, the avatar suggested adding disclaimers to anything that could be misread as a Meta endorsement. The framing differs from competitors' voice assistants: Meta is selling companionship rather than answer latency.
Realtime conversations currently work on mobile, web and Mac, with Meta saying AI Glasses and the upcoming Muse Charm device will follow. That Tamagotchi-like wearable is slated to arrive before year-end, with the avatar appearing on a small screen so users can talk to it and hand it tasks.
The timing is not accidental. Muse has roughly 10 million users today, and Meta wants consumer AI to become a business serving billions; shipping realtime voice and video ahead of the dedicated hardware pre-builds the interaction pattern the device will depend on.
The real test is retention and trust. An avatar that blinks and gestures will likely get people to say more than a text box would, but it also drags privacy, use by minors and emotional dependence into the foreground. Watch the Muse Charm launch experience, and whether Meta opens realtime conversations to third-party developers.
Why it matters
Making voice and video the default interface shifts consumer AI competition from raw model quality toward companionship and hardware entry points, while forcing privacy and emotional-dependence questions into the open.
Nearby Updates
All11/06, 03:55
Cowboy raises $5M seed to build an AI agent marketplace and developer platform
Cowboy has raised $5 million in seed funding to launch an AI agent marketplace and developer platform, Pulse 2.0 reported. The round is aimed at letting agents be built, listed and distributed like applications, though investors and product details have not been disclosed.
11/06, 04:00
NVIDIA publishes Vera CPU benchmarks: 1.6x more agents per watt
NVIDIA says its Vera CPU beats both the fastest and the most energy-efficient x86 chips on agent workloads, claiming 1.6x more agents per watt and 2x faster agent speed in a 100-megawatt data center comparison. Partners including Perplexity, DeepInfra, ClickHouse and Daytona report production-style gains, though every figure comes from NVIDIA or its partners.
11/06, 03:35
NVIDIA open-sources four Cosmos 3 world models for robot and driving simulation
NVIDIA released four open-weight Cosmos 3 checkpoints for interactive world simulation, covering multi-sensor driving data generation, streaming driving simulation, simulator-control-to-video rendering and bimanual robot action prediction. All four are post-trained from Cosmos 3 Nano with 16B trainable parameters and ship under the OpenMDW 1.1 license alongside a technical report and post-training recipes.
11/06, 04:03
METR Analysis: Anthropic's New Model Clears 50% on Every Benchmark
A METR analysis found Anthropic's newest model scoring above 50% on every benchmark in the evaluation set, Crypto Briefing reported on November 5. The result is a breadth signal rather than a leaderboard win, and per-benchmark detail still has to come from METR's underlying report.