Nvidia and Meta announced a strategic partnership in which Nvidia plans to invest $3 billion in Meta. Meta will expand its deployment of Nvidia's next-generation infrastructure, including the Vera Rubin architecture, and the companies will jointly optimize the Muse large model's inference and training performance to lower total cost of ownership.
During its Q3 2026 earnings call, Meta disclosed that its AI app Muse has surpassed 10 million users. CEO Mark Zuckerberg said about 25% of Muse users use it daily, and more than a quarter of active shopping users spend over $200 on average; he also cited user trust in Muse's security setup and said Meta will focus on scaling AI agents, increasing revenue, and optimizing compute costs, expecting Muse's revenue per user to exceed that of its family of apps.
Meta announced a new business line selling AI tools to enterprise customers and restructured its organization around the consumer AI assistant Muse. Muse has surpassed 10 million downloads since its September launch, with subscription plans at $20 and $100 per month.
According to Sensor Tower, Meta's AI assistant Muse has reached 10.9 million downloads since its September launch, including 10.2 million in the U.S., making it one of only 12 apps to hit the milestone within 40 days. Meta drove growth through cross-app marketing and advertising, with Muse accounting for 39% of its ad spend in October.
Meta announced a terms of service update effective October 30, authorizing Meta and its agents to use automated tools such as web scrapers to crawl, index, and cache content from websites associated with business users' products and services, for use in Muse AI agent browsing, advertising, and model training. Critics say it could help Meta avoid tens of millions of dollars in content licensing fees and potentially restrict news organizations' access to its platforms through contractual conditions.
Meta Platforms announced the release of open-weight model Muse Spark Open, allowing developers to run and adapt model parameters. The model is designed for coding and AI agent support, and is less advanced than Muse Wonder, which was unveiled earlier this week. Meta's chief AI officer Alexandr Wang said the company usually keeps its most powerful models closed for safety and capability reasons, but will continue to open-source strong models.
Meta's Superintelligence Lab released Muse Wonder and Muse Wonder Full, two frontier models emphasizing long-term memory, with benchmark scores comparable to Fable 5.1 and GPT-6 Astra. Wonder is available via the Meta Model API at $4 per million input tokens and $20 per million output tokens, while Wonder Full is exclusive to Muse Code subscribers.
Alexandr Wang, Meta's chief AI officer, said Meta will continue to improve Muse's memory and complex task handling, reminding users that they need to make efforts themselves for the agent to earn money. Muse remains the top free app on Apple's App Store one month after launch. Meta offers tiered subscriptions and has not spent much time on ads, but may charge fees from user transactions such as shopping, saving, and earning money.
Bi Shuchao announced that Meta's Muse Spark 1.1 and 1.2 models helped mathematicians publish six papers in six fields over six months, answering five open problems and disproving the Kida conjecture in group theory with a 384-order semiabelian non-monomial group.
Meta officially launched Muse Gadgets, opening up its hardware capabilities so developers can run Muse firmware on ESP32 boards or use the Linux SDK to connect devices such as the Raspberry Pi 5 to build custom Muse peripherals. It also introduced the official Muse Home Link device, with an initial batch of 5,000 units expected to ship within weeks and offered free to eligible Muse subscribers while stock lasts.