Meta's Muse Passes ChatGPT: 730,000 Downloads and the System Behind It
Meta's Muse app overtook ChatGPT to become the No. 1 free iOS app in the U.S., with 730,000 downloads in its first five days. The figures were reported on Sept. 22, placing Muse ahead on Apple's U.S. storefront in less than a week of availability Yahoo Finance.
The ranking follows Meta's Sept. 8 announcement titled "Introducing Muse: The World's First Personal AI Agent Built for ..." Meta calls Muse its personal AI agent (software that does tasks for you, not just answers) and describes it as secure and private. The company stated Muse is rolling out in the U.S. on iOS and Android Meta.
Muse runs on models built under Meta Superintelligence Labs. Meta describes Muse Spark as a natively multimodal reasoning model, meaning it can handle text, images and other inputs together, with support for tool-use, visual chain of thought and multi-agent orchestration Meta. Like a manager splitting a job among specialists. The company stated Muse Spark powers a smarter and faster Meta AI assistant and will roll out to WhatsApp, Instagram, Facebook and Messenger.
Meta introduced Muse Spark 1.1 as a multimodal reasoning model built for agentic tasks, or work done on a user's behalf. Powered by Muse Spark 1.1, Meta AI can make plans, connect to email and calendar apps, create slides and handle tasks for users, according to the company. Meta describes Meta AI as its free personal AI chatbot and assistant.
The company launched the first version of the Meta AI app as an assistant that gets to know preferences and remembers context. That app was introduced on April 29, 2025. Muse extends that idea from remembering to acting across systems and third-party connections.
On media generation, Meta announced the launch of Muse Image and a preview of Muse Video. The company states Muse Image and Muse Video are the first media generation models developed by Meta Superintelligence Labs. Meta stated Muse Image is available in Meta AI.
Reuters analysis found that Meta's Muse Image failed to identify some of its own AI-generated images once they were cropped. The finding was published July 11 and concerned the detector behavior on cropped outputs Reuters.
The broader context here is speed versus staying power. A No. 1 rank and 730,000 installs in five days measures early demand. It does not measure retention, daily active use or compute cost per session. The test is how many installs link email, calendar and other lasting sources, and how often tasks finish without help.
In my view, the design is what to track. Tool-use, visual chain of thought and multi-agent orchestration point to a system that splits work across smaller helpers rather than doing it in one pass. That can help with multi-step jobs. It also adds failure points around permissions, tracking work and keeping records. The Reuters result on cropped images is a separate warning. Proof of origin fades fast once pixels leave the original app, and that matters for trust and safety if creation and sharing live together.
