Meta has made its ambitions in artificial intelligence unmistakably clear. The company has officially introduced Muse Wonder and Muse Wonder Full, the first models it classifies as "frontier models" — the most advanced tier of AI systems.

Thanks to their massive scale, the new models are designed to deliver strong performance on complex tasks and coding work in particular.

What the Wonder series promises

museic

According to Meta's internal testing, the Wonder series produces results on par with powerful rivals such as Fable 5.1 and GPT-6 Astra. In an official blog post, the company described the models as the next major step in its scaling journey and a significant leap toward its vision of personal superintelligence.

How pricing will work

That enormous scale naturally comes at a cost. The standard Wonder model, accessible through the Meta Model API, will be priced at $4 per million input tokens and $20 per million output tokens. The strongest member of the family, Muse Wonder Full, will not be available via the API at all; it will be reserved exclusively for Muse Code subscribers.

Both models will power the company's AI assistant, Muse.

Active memory takes center stage

The models' most striking capability is a feature called "active memory". It allows the AI to store and synthesize information received from the user over long periods of time.

Meta says it has invested heavily in this area so that the models can remember details the user considers important even across conversations spanning days, weeks or even months.