For several years, Meta has talked a big game about artificial intelligence while trailing the companies that set the pace. This week it says that gap has closed. On 13 October 2026, Mark Zuckerberg unveiled two new models, Muse Wonder and Muse Wonder Full, which Meta claims are as capable as the latest systems from OpenAI and Anthropic. As The Independent reported, the company describes the larger of the two as the "largest and most capable model we have trained".
Benchmarks and bragging rights aside, the more interesting part of the announcement is what Meta wants these models to do. It isn't pitching a better chatbot. It's pitching an assistant that takes on long, messy, real-life jobs for you. This piece explains what was announced, how it differs from what came before, and what to think about before you hand an AI agent a piece of your life.
What Meta actually announced
There are two models. Muse Wonder is the general-purpose one, meant to handle most everyday tasks. Muse Wonder Full is the heavyweight, aimed at harder problems. Both build on work from Meta's Superintelligence Labs division, which released a free "Meta Muse" agent last month. Meta says the new models will soon power more of that agent.
Zuckerberg called the launch a step towards "personal superintelligence for everyone". That phrase sums up how Meta's strategy differs from its rivals'. Other labs tend to talk about coding, research or enterprise productivity. Meta is putting the emphasis on helping individuals manage their own lives.
The headline features
Long, multi-day tasks
Meta gave two telling examples of the work it has in mind: helping with "a cross-country move" and dealing with "a claim their provider denied". Neither is a single question. Each takes dozens of steps spread over days or weeks: gathering documents, comparing options, filling in forms, following up and reacting when something goes wrong. Less capable models tend to make compounding errors over that kind of time span, and Meta says the new models are built to hold up better.
Active memory
Meta says the models have "active memory", which lets them keep track of a task over time, remember the context that matters and pull up the right details when they're needed. For anything that unfolds across several sessions, this is the feature that counts. An assistant that forgets which quote you rejected last Tuesday isn't much of an assistant.
Staying on task, but checking in
Meta also says the models are better at keeping their focus and continuing towards a goal when they hit obstacles. It is careful to balance that against restraint, saying the models will "ask for clarification when needed" and check before "taking important actions". It also says they resist "adversarial instructions" better, meaning text designed to trick a model into misbehaving.
What it costs
According to The Independent, using the new models through Meta's paid access is notably more expensive than Meta's earlier systems, though in line with top-tier models from other labs. For most people, the more relevant point is that Meta says the free Muse agent will increasingly run on these models. You may end up using them without ever choosing to.
Why this launch matters beyond Meta
The announcement came in a busy week. On the same day, Anthropic published research warning that small AI models are getting closer to being able to copy themselves across poorly secured home devices, and the industry newsletter Tech Bytes noted a cluster of other AI infrastructure launches. The direction of travel is clear. AI is moving from answering questions to acting on our behalf, and the companies building it are competing over who can be trusted to do that.
That raises the stakes. If a chatbot gives you a bad answer, you can ignore it. If an agent files the wrong form, emails the wrong person or agrees to the wrong quote, the mistake has consequences in the real world.
The useful question isn't "Is this model smart?" It's "What am I comfortable letting it do without me?"
Practical questions before you delegate to an AI agent
Whether you end up trying Muse Wonder or a rival, these habits will serve you well:
- Start with low-stakes tasks. Use an agent to research removal companies or draft a complaint letter before you let it book or submit anything.
- Check the confirmation settings. Meta says its models will check before important actions. Find out where that setting lives and keep it switched on.
- Review what it remembers. Memory makes agents useful, but it also means personal details are kept. Look for controls that let you see and delete stored context.
- Limit account access. Only connect the email, calendar or payment accounts the task actually needs, and remove access when the job is done.
- Keep your own paper trail. For anything involving money, insurance or legal deadlines, save copies of what was sent and when.
What we still don't know
Several things are unclear. Meta's claim that it now matches its rivals will need independent testing. It's also too early to say how the models handle sensitive personal information in everyday use, or how often the promised "check before acting" behaviour kicks in. And Meta's business has always relied on advertising, so people will reasonably ask how data gathered by a personal agent will and won't be used. That deserves a clear public answer.
The bottom line
Muse Wonder marks Meta's arrival at the front of the AI race, by its own account at least, and a bet that the next big consumer product is an assistant that handles life admin for you. The promise is real: tedious, multi-step chores like moving house or challenging a rejected claim are exactly where a patient, well-organised helper could save hours. But the more an assistant can do, the more carefully you should decide what it's allowed to do. Try it on small tasks first, keep the guardrails on and widen its role only as it earns your trust.
