Meta's Muse Agent Hits 5M US Downloads in 22 Days, Faster Than ChatGPT

Meta's personal AI agent Muse, released on September 8, gained about 5M downloads in the US in roughly 22 days. The estimate comes from Sensor Tower. According to it, Muse is growing faster than ChatGPT, Grok and Claude did over a comparable period after launch, and it has already topped the US app store rankings. For a company that was recently criticized for lagging behind OpenAI and Google, this is a notable change in tone.

The source, the Chinese outlet Jiqizhixin, tells the story through an interview with Alexandr Wang, head of Meta's AI division. The headline is a three-part question: why Muse became a hit, whether the expensive hiring was justified, and what Wang himself has to say. The excerpt available to us cuts off before the answers, so there are no Wang quotes here, and we are not guessing at them.

The numbers we know

Sensor Tower's estimates are built on app download data, not on Meta's own reporting. The historical comparison with ChatGPT, Grok and Claude also relies on retrospective estimates, not official company figures. The precision of such comparisons is limited. But the order of magnitude is clear: 5M downloads in a little over three weeks is a lot for an agent product.

• Product: Meta Muse, a personal AI agent
• Launch: September 8
• Trajectory: about 5M US downloads in 22 days
• Source of the estimate: Sensor Tower
• Comparison: faster than ChatGPT, Grok and Claude, based on historical estimates
• Market: the US, where Muse climbed to the top of the chart

Context: the "lobster" has cooled, and

Meta is back in the conversation

The first part of the headline, "龙虾凉了", literally means "the lobster has cooled." In the Chinese tech press, "lobster" apparently referred to the recent wave of interest in open agents, and that hype has now died down. Against that backdrop, Muse looks like a product that caught the audience's attention just as the previous wave receded.

The second storyline is about people. In 2025, Meta invested about $14.3B in Scale AI and brought in Wang to lead its superintelligence research. At the same time, the company reportedly offered researchers record-setting pay packages to assemble a new team. At the time, it looked like a bet with an uncertain payoff. Now Wang has an argument in the form of a public metric, and the question of what the hiring cost is being discussed with a working product in the picture.

What has not been proven yet

Downloads are not the same as retention. An app can rack up millions of installs in its first month and lose most of its audience by the second. There is no data on active users, on how often people use the agent, or on revenue. It is also unclear how much of Muse's success comes from the model itself and how much from distribution through Meta's ecosystem, where the company has billions of users. That last factor can explain the rapid ramp-up without any technological superiority.

What to watch next

If you follow the agent market or are building a product near it, check three things. First: whether Sensor Tower or similar firms publish 30- and 60-day retention data for Muse. Second: whether Muse expands beyond the US and how it performs where Meta's position is weaker. Third: what Wang and Meta disclose about the model under the hood and the cost of inference at this load. Only then will it be possible to judge whether the money spent on the team paid off. It is worth reading the full interview in the original before drawing conclusions about his arguments.