Meta’s pocket AI device could cost $300
Muse already works in apps, WhatsApp and on the web. Charm would add hardware, not a new AI agent.
By Saeed Ezzati8 min read

The audio edition
Listen to this newsletter
0:005:13
Read transcript
Meta’s pocket-size AI device, Muse Charm, could cost $300—but that price is reported, not confirmed. The Information reported Meta intends to charge that amount, in coverage by TechCrunch. Asked about the figure, Meta didn’t confirm it. A spokesperson said the company would have more to share soon, without naming a date. Charm is designed as a dedicated way to talk with Meta’s Muse agent, using a real-time voice model. Meta introduced the device at Connect in September and described it as small enough to fit in a pocket. The important distinction: Charm would give Muse its own hardware, not introduce a new AI agent. Muse is already available through Meta’s app, WhatsApp, the web and a Mac desktop version. TechCrunch describes Charm as Tamagotchi-like—a physical presence for software people can carry and speak to. At the reported $300 price, Charm would sit above Friend’s $250 wearable, which supports text and voice interaction with an AI persona. Amazon’s Bee is a lower-cost comparison with a different focus: TechCrunch listed its regular hardware price at $75. Bee transcribes conversations and tracks tasks, and TechCrunch also cited a $99 annual subscription for connecting it to apps such as photos and email. These products aren’t interchangeable, but they show the range of prices and jobs emerging in AI wearables. For potential buyers, the next milestone is information, not a confirmed launch or price. At Connect, Meta said it would share more about Charm later in 2026. Its latest response still leaves the timing, final pricing and launch details unresolved. The practical question is whether dedicated hardware makes an already-available agent useful enough to justify its cost. That gap between a headline number and what it actually tells you showed up in Anthropic’s first day on the public market. Shares opened on Nasdaq at $111.50, against an IPO price of $83—about 34 percent higher. By 1:55 p.m. New York time, they had slipped to $95.69. That was still above the offering price, but well below the opening trade; it was an intraday quote, not the closing price. The expanded sale of 980 million shares at $83 implied about $81.3 billion in proceeds for Anthropic and employees together—not all for the company. Separately, the IPO, existing fundraising and new private offerings timed to the debut are projected to bring Anthropic’s cash reserves to $220 billion. Those are distinct figures. CEO Dario Amodei said the stronger financial position would not mean relaxing safety constraints, pledging to pace training and deployment until the company has confidence its systems are safe and remain under human control. The opening surge, then retreat, is a reminder that one early quote is not a settled market verdict. And beyond markets, AI’s value is also being tested against real-world outcomes. A peer-reviewed University at Albany study examined Google’s WeatherNext Cyclone model across 385 tropical cyclones worldwide, from June 2022 through May 2026. For storm strength, it beat NOAA’s hurricane model at every forecast lead time from one to five days. Beyond the first day, it performed about on par with human forecasters. The caveat matters: the model tended to underestimate intensity, increasingly so at longer lead times, and its ranges were often too narrow to include the actual storm strength. Performance also varied by region. Researchers said it was most accurate in the North Atlantic and weakest in the eastern North Pacific. The evaluation does not cover a newer Google version released in 2026, so it says nothing about that release’s performance. The researchers see AI as another tool for forecasters, not a replacement. Finally, the broader market showed why Anthropic’s strong debut shouldn’t be read as a lift for the whole AI trade. On November 19, the S&P 500 fell 2.03 percent, extending its losing streak to five sessions. Preliminary figures put the Nasdaq’s decline at 2.23 percent. Nvidia fell for a sixth straight session despite an upbeat revenue forecast, while chipmakers and cloud-computing providers CoreWeave and Nebius also came under pressure. Argent Capital portfolio manager Jed Ellerbroek cautioned against treating the day as a simple verdict on AI, noting that investors making room for Anthropic shares may have sold other holdings. He also said his expectations for data-center spending growth had eased, with projects potentially becoming more expensive and slower to build. Meanwhile, markets priced in about a 94 percent chance of a quarter-point Federal Reserve rate increase in December. That was an expectation, not a decision. Across these stories, the useful thing to watch is what holds up beyond the first headline: whether buyers see enough value to pay for dedicated AI hardware, whether Anthropic’s market price stabilizes, and whether models and markets deliver on their promises once the caveats are counted.


