Meta's Muse Hits 10M U.S. Downloads in 40 Days: What the Purchase Data Shows
Meta's Muse agent passed 10.2 million U.S. downloads in 40 days, with Q3 earnings revealing a $200+ average basket. Retention, pricing, Amazon's block and Walmart's Agent Gateway, by the numbers.
Meta's Muse hit 10 million U.S. downloads in 40 days — and the purchase data is the real story
Meta's personal AI agent Muse, launched in the U.S. and Canada on September 8, 2026, passed 10.9 million downloads by October 20, with 10.2 million of those in the United States, according to Sensor Tower. Only 11 other apps have ever reached 10 million U.S. downloads within 40 days of launch. The most recent was Meta's own Threads in July 2023. Most of the rest are mobile games — Fortnite, PUBG Mobile, Pokémon GO. Outside gaming, the list contains just Threads and Disney+.
On October 21, during Meta's third-quarter earnings call, Mark Zuckerberg confirmed the figure himself: more than 10 million users. Taken together, the download milestone and the earnings disclosure are the first hard evidence that consumer AI agents are moving real money, not just generating installs.
Putting the numbers in order
What Sensor Tower's data actually shows:
- Downloads: 10.9 million in 40 days, 10.2 million of them U.S. (roughly 94%)
- U.S. comparison at 40 days: ChatGPT 3.7 million, Grok 500,000, Claude 200,000
- Global comparison: ChatGPT did about 15 million in its first 40 days, ahead of Muse — but ChatGPT was available far more widely, and Japan alone accounted for 25% of its early installs. Muse is U.S. and Canada only.
- Daily active users: 2.5 million for the first time on October 16, up 15% week over week
- Retention: 30% at day 7 and 23% at day 14, versus a 19% / 16% average across ChatGPT, Claude and Grok
Meta is also buying part of this. The company put 39% of its total ad spend behind Muse in October, up from 23% in September, and climbed from 73rd to 21st among U.S. brands by ad impressions. A company with 3.6 billion daily users across its apps can manufacture a chart-topping launch. It cannot manufacture a 7-day retention rate that beats three rivals by 11 points.
What the earnings call added
The numbers that matter appeared on the call, not in the download chart:
- About 25% of people who set up Muse now use it daily.
- More than a quarter of engaged users who shopped through the agent had an average purchase size above $200.
- Plaid, the fintech that connects the agent to bank accounts, told Meta it "hadn't seen activity like they're seeing with their connector on any other service before."
- Beyond shopping, the top use cases are email and calendar productivity tasks.
A $200+ average basket is not impulse buying. It looks like appliances, furniture and bulk orders — the kind of purchase where delegating the research-and-checkout loop is worth something. That is a meaningful signal: if agent-mediated commerce concentrates in high-ticket categories, the economics of affiliate and checkout fees change accordingly.
The pricing model, layer by layer
Muse is free up to a weekly allowance of 100 million tokens. Two paid tiers sit above it: Power at $20/month for 500 million tokens a week, and Maximum at $100/month for 3 billion.
Meta has said it intends to take a small fee on transactions over time, but has published no rate, no start date and no payer. Shopify's documentation lists no Meta-specific fee beyond standard payment processing today. For context, OpenAI charged merchants 4% on sales completed inside ChatGPT's Instant Checkout before retiring the product in March 2026; Google and Microsoft charge nothing extra for agent checkouts.
Analyst math on the fee is speculative by nature. Deutsche Bank projects up to $36 billion in annual revenue by 2030, with a low end of $2.4 billion. Hitting $36 billion through subscriptions alone at $20/month would require roughly 150 million paying users against today's 2.5 million daily actives. The transaction-fee path is the plausible one — and it depends entirely on merchant permission, which is exactly where the fight is.
Merchant permission is the bottleneck
Amazon blocked Muse from its store on September 20, saying the agent did not identify itself and that Amazon had not agreed to participate. JPMorgan Chase confirmed on October 13 that it restricts outside agents from obtaining banking credentials to move customer money, though read-only balance access remains available. Travel and booking sites have locked agents out entirely.
Walmart went the other direction on October 21, announcing an MCP-based Agent Gateway and Agent Profiles rolling out next month. Each agent gets a dedicated profile in the customer's account, purchases are attributed separately from the customer's own, and shoppers set either per-transaction approval or a spending cap — the same model as adding an authorized user to a credit card. Walmart also published terms for agents: they cannot exceed the shopper's intent, cannot access payment details, and get only the data needed for the task. Non-compliant agents are blocked. Walmart's standard return policy applies even when the agent misbehaves.
That last clause is the one most businesses will copy. It moves the risk of a malfunctioning agent off the customer and onto the retailer, which is the minimum condition for mainstream adoption.
Where this leaves builders
Muse is not available in Japan or Europe yet, and two documented incidents show why a cautious rollout makes sense: reports that Muse read messages users had not authorized, and a Facebook Marketplace case where an agent negotiated a lower price and disclosed the seller's home address to a buyer, who then showed up at the house. Permission design, not model quality, is the limiting factor.
The practical takeaway for anyone building in this space:
The download record will be forgotten in a month. The $200 average order and the authorized-user model for agents are the parts that will still be shaping how commerce works next year.