Meta’s newest artificial intelligence is built to remember what matters to you for weeks and months at a time, then keep working on a job until it gets done.
Meta Platforms released Muse Wonder and a larger sibling, Muse Wonder Full, on Monday, calling them its most capable models and saying they lead on tasks where an AI agent acts on a person’s behalf at work and at home. They are the company’s first entry into the “frontier” tier, the top class of AI models now dominated by OpenAI and Anthropic. Meta described the release as “a major leap toward personal superintelligence.” meta
In plain terms, Meta’s earlier model, Muse Spark, was built for speed and quick answers. Wonder is the bigger, heavier-duty version, aimed at harder jobs like writing software. Meta says it was also trained on everyday problems in law, finance and health, from reading a lease to making sense of financial or medical information. TechCrunch meta
The feature Meta is pushing hardest is memory. The company calls it Active Memory: the model carries important details about you from one conversation to the next and pulls them up when they become relevant to the task in front of it. So if you mentioned a budget, a deadline or a food allergy two weeks ago, the idea is that you won’t have to repeat yourself. meta
The second feature, Active Focus, is about follow-through. When the model hits a roadblock on a hard task, Meta says it tries other approaches, adapts to new information, asks for clarification when needed and checks with the user before taking any important step. meta
On performance, Meta’s own testing puts the two Wonder models roughly even with the top current offerings from Anthropic and OpenAI, and ahead of them on a test that measures how well an AI remembers and pieces together what users tell it. Those are company numbers, and outside testers have not yet published their own results. TechCrunch
Who can use it right now comes down to two groups. Muse Wonder is available today to developers through the Meta Model API and Meta’s coding tool, Muse Code, while Wonder Full is open only to Muse Code subscribers. Wonder Full will not be offered to outside businesses through Meta’s developer service. meta TechCrunch
Regular consumers will have to wait a little longer. Meta says Wonder will soon power its Muse personal assistant app, with paying subscribers getting it first. The company also plans to use Wonder to build smaller, specialized models that it can bring to everyone using Muse. meta
For businesses that want to build their own apps on top of it, the price tag puts Meta in the same league as its rivals. TechCrunch reported that Wonder will cost $4 for every million tokens of text sent in and $20 for every million tokens it writes back, roughly matching what competitors charge for their premium models. A million tokens works out to about 750,000 words, or around eight full-length novels. Put simply, having the AI write costs five times as much as having it read. Meta is choosing to compete on quality, not on undercutting rivals on price. TechCrunch
The model was built on American soil. Meta trained both Wonder models at Prometheus, its gigawatt-scale data center, which sits in central Ohio and came online earlier this year. A gigawatt is the kind of power output that normally comes from a large power plant, which shows how much electricity top-tier AI now demands. meta TechCrunch
The launch caps a fast rebuild. Meta says it began a ground-up overhaul of its AI efforts fourteen months ago. That push is led by Alexandr Wang, the Scale AI founder who has been Meta’s Chief AI Officer since 2025 and runs Meta Superintelligence Labs. The first result, Muse Spark, arrived in April and was pitched as closing the gap with OpenAI and Anthropic. Meta then opened its Muse personal AI agent app to the public on September 8. Until Monday, though, Meta had stayed away from the kind of top-of-the-line model releases its biggest rivals are known for. Muse Wonder: A Frontier Model For Personal \& Professional AI Agents +4
An assistant that remembers your life and takes action for you also raises an obvious question about privacy. Meta acknowledges that a personal agent may handle messages, calendars and private documents, and says it trained Wonder to be discreet with personal information and to resist instructions designed to hijack it. The company says the model also carries its strongest protections yet against being misused to help with chemical, biological, radiological or nuclear weapons or cyberattacks, while still allowing defensive security work. A full report with more safety testing is due soon. meta meta
For everyday users, the takeaway is simple: Meta is betting that the AI people keep coming back to will be the one that remembers them. The test begins once Wonder reaches the Muse app and ordinary customers can see whether it actually does.
JBizNews Desk | Menlo Park, Calif.
© JBizNews.com All Rights Reserved. Reproduction or distribution without written permission is prohibited.
