Meta has released Muse Wonder, the model it now calls its most capable, and it is built around two ideas: remembering what matters to you and not giving up on a task. The series has two versions, Muse Wonder and the larger Muse Wonder Full, according to Meta.
Meta pitches it as a step towards what it calls personal superintelligence. In plainer terms, it is the engine for agents that book, plan and build things on your behalf, which is where Meta, OpenAI, Anthropic and Google are all now competing.
What Meta’s new model actually does
Meta groups the new abilities under two names. Active Memory carries important context about you across conversations and pulls out the details that matter for the task in front of it. Active Focus is about persistence: trying another approach when one fails, working through blockers, asking for clarification and checking in before important actions.
Meta’s demos show the model building a playable 3D game from images, planning a weekend run in a browser and helping with an insurance claim. It says both models were trained at its Prometheus data centre and lead on benchmarks for personal and professional agents, memory and coding, with domain tasks in law, finance and health.
Where you can use it today
Muse Wonder is available now through the Meta Model API and Muse Code, Meta’s coding tool. Muse Wonder Full is limited to Muse Code subscribers. Meta says Muse Wonder will soon power its Muse app, rolling out to paying subscribers first.
Meta also makes a broader claim. Comparing Muse Wonder Full with its earlier Muse Spark 1.3, it says the new model does better than expected on tasks it was never specifically trained for, because it sits on a larger base model. That matters for agents, which face combinations of goals and tools nobody planned for.
There are gaps. Meta’s post does not publish prices, benchmark scores or a list of countries, and it says a full safety report is coming later. Until those arrive, every claim about how it compares with Claude or GPT comes from Meta alone.

What it means for Australians
The Muse app still is not open in Australia. As we covered in our Meta Muse explainer, it works in the United States, Canada and Mexico, and Meta has named no date for Australia. Muse Wonder does not change that, so most Australians will meet it later, inside the app.
Memory is also where the risk sits. An assistant that remembers details across chats holds more personal information, and Muse already drew criticism last week over how it handles messages. Australian businesses testing it should treat that context as data covered by the Privacy Act 1988.
Why Australian businesses should care
Agents that remember customers and finish tasks without hand holding are exactly what many firms want from AI. But memory changes the questions to ask a vendor: what is stored, for how long, where, and who can see it. Those answers matter more than a benchmark chart.
If you are weighing an agent for customer service or internal work, start with a narrow task, keep a person approving important actions and check the data terms first. It is the same approach we take in our AI chatbot development work.
The race for agents that remember
Fourteen months after Meta says it began rebuilding its AI work from the ground up, it is now pitching a frontier model rather than catching up. Whether Muse Wonder lives up to that will depend on independent tests and the safety report Meta has promised. For Australians, the bigger question is still when Muse arrives here at all.
