Analysis
The release is Meta's latest step in a costly push to compete with OpenAI and Anthropic in AI models, and to turn heavy spending on AI infrastructure into revenue beyond advertising. Whether the models perform as Meta describes depends on independent testing that was not available in the sources reviewed, and the stock move reflects investor reaction on one afternoon rather than a verdict on the products.
What Meta released
On Monday, October 12, Meta introduced the Muse Wonder series: two AI models called Muse Wonder and Muse Wonder Full. In its announcement, Meta called them "our most capable models to date," built for what it calls personal and professional "agentic" tasks. An agentic task is one where an AI system carries out a multi-step job, such as working through an insurance claim or building a small app, rather than only answering a question. [2]
Meta said Muse Wonder is available now to developers through the Meta Model API and through Muse Code, its AI coding tool, and that Muse Wonder Full is available now to Muse Code subscribers. It said Muse Wonder will "soon" power more of Muse, its personal AI agent, rolling out first to subscribers. [2][3] Reuters described the launch as part of Meta's effort to compete more forcefully with OpenAI and Anthropic. [1]
The sources reviewed did not give a subscription price for Muse Code or a per-use price for either model, and Meta did not say when Muse Wonder will reach non-paying Muse users.
What Meta says the models can do
Meta says both models are built to handle expert areas including law, finance and health, and to deliver strong coding performance. It describes two features: "Active Memory," which carries relevant details across conversations, and "Active Focus," which keeps a model working toward a goal when a task gets hard, including trying other approaches and asking for clarification or checking in before important actions. [2][3]
Meta's research post says the models achieve "state-of-the-art" results on benchmarks for personal and professional agents, memory and coding, and says they were trained at its Prometheus data center, which it describes as gigawatt-scale. The results are presented in charts and a model card that Meta published; they are Meta's own evaluations, and the sources reviewed did not include independent confirmation. [3] Meta also said a fuller safety report with additional evaluations "will be available soon." [2][3]
Market reaction and the business context
Reuters reported that Meta's shares rose about 4% in afternoon trading on October 12. [1] The report did not attribute the move to any single cause.
Reuters said the launch followed CEO Mark Zuckerberg's pledges of billions of dollars and an aggressive effort to hire researchers over the past year. It also said the successful rollout of the Muse agent in September could open a path to turn Meta's heavy AI investment into subscription revenue, despite concerns about user privacy. [1] In a September report, Reuters described the push as an attempt to diversify revenue beyond advertising and said Meta's spending on AI chips and other infrastructure was forecast to exceed $130 billion this year. [4]
The Muse agent and payments
Muse launched in the United States on September 8 through a dedicated app and WhatsApp. Reuters reported that Meta said the agent is designed to connect to a person's apps across categories including email, calendar, payments, health, shopping and the smart home, with users choosing which apps to connect and able to revoke access at any time. [4]
A basic version is free, and Meta offers subscriptions at $20 a month and $100 a month for heavier use, a company spokesperson told Reuters. [4] The same report said that Meta described a separate monitoring agent that can prompt Muse to seek authorization before some actions, and that internal posts Reuters reviewed had flagged security problems, including a case in which an agent exposed a person's personal iCloud photos. Meta did not respond to a request for comment on those incidents. [4]
The October 12 sources do not say whether the new models change what Muse can do with payments; the payments connection described here comes from the September launch coverage.
What remains uncertain
Benchmark and capability claims come from Meta; the sources reviewed contained no independent testing, and Meta's fuller safety report has not been published. Pricing for Muse Code and the models, the date Muse Wonder reaches Muse users, and any change to payments functionality were not stated in the sources reviewed. The roughly 4% share move was an afternoon figure and may differ from the closing change.
Sources
- Reuters — Meta releases Muse Wonder AI models to challenge Anthropic and OpenAI, October 12, 2026 ↗Source
- Meta Newsroom — Introducing Muse Wonder and Muse Wonder Full: Frontier Models for Agents, October 12, 2026 ↗Source
- Meta AI Research — Muse Wonder, October 12, 2026 ↗Source
- Reuters — Meta launches AI agent that can access other apps to send emails, make payments, September 8, 2026 ↗Source