Meta launches Muse Wonder frontier models for personal and professional AI agents

Meta on Monday introduced Muse Wonder and Muse Wonder Full, describing them as its most capable AI models yet and its first clear bid at the frontier class dominated by OpenAI and Anthropic. In a Meta Newsroom post, the company said the models deliver “state-of-the-art performance on personal and professional agentic tasks,” from navigating an insurance claim to reviewing a lease or building a working app. Muse Wonder is available immediately through the Meta Model API and Muse Code; Muse Wonder Full is available to Muse Code subscribers.

A companion Meta research blog said the release builds on Muse Spark and was trained at Meta’s gigawatt-scale Prometheus data centre. Meta highlighted two capabilities it calls Active Memory — carrying important context about a user across conversations — and Active Focus — continuing toward a goal when tasks get hard, trying other approaches, recovering from errors and checking in before important actions. “Together, they mark the next step on our scaling ladder and a major leap toward personal superintelligence,” the research post said.

TechCrunch reported that Meta’s own testing, presented in a model card, showed Wonder and Wonder Full achieving scores comparable to Fable 5.1 and GPT-6 Astra. API pricing for Wonder is US$4 per million input tokens and US$20 per million output tokens, roughly in line with GPT-6 Sol or Opus 5.5, TechCrunch said. Wonder Full will not be served through the public API. Both models are also slated to power Meta’s Muse agent, rolling out first to subscribers.

On CNBC, Meta AI chief Alexandr Wang said Muse Wonder — previously known internally as Watermelon — was trained so agents do not stall on everyday tasks such as booking dinner reservations. “When you do real-world tasks, you’re going to get roadblocks and you’re going to get things that come up or various setbacks,” Wang said. “We specifically tried to train the model to be able to try other approaches instead of stalling and then checking back in with the user at any time.”

CNBC said access to Wonder features inside the Muse consumer app will require a monthly subscription of US$20 or US$100, underscoring Meta’s push for AI revenue beyond advertising. An even more powerful Wonder Full tier for the Muse Code coding agent is offered under US$50 or US$200 monthly plans, according to CNBC. Wang called the API offering a much larger model at frontier level, while saying it would still be priced meaningfully below leading OpenAI and Anthropic offerings.

Meta said Muse Wonder “offers our strongest safeguards to date” against cyber misuse and was trained to avoid sharing private information without permission and to refuse help with harmful acts, with a fuller safety report promised soon. TechCrunch noted it is the first frontier-model launch since Scale AI founder Alexandr Wang restructured Meta’s AI organisation in August 2025, after the company had shipped the lighter Muse Spark model and the Muse agent without matching rivals’ top-tier releases.

Wang told CNBC the company is monitoring agent-related incidents and educating users after some social-media complaints that Muse accessed personal messages — claims Meta has disputed — “to ensure that consumers can trust the product going forward.” Meta said Muse Wonder will also power artifact generation and higher-intelligence processes inside Muse, and that specialized models derived from Wonder are planned for broader Muse users later.

Sources: Meta Newsroom: Introducing Muse Wonder and Muse Wonder Full (Oct. 12, 2026); Meta AI research blog: Muse Wonder (Oct. 12, 2026); TechCrunch (Oct. 12, 2026); CNBC (Oct. 12, 2026).