Meta's first frontier model, Muse Wonder, takes on Fable 5.1

Meta has spent the time since Alexandr Wang restructured its AI efforts in August of last year on smaller releases — the lightweight Muse Spark model and the Muse agent — rather than the kind of state-of-the-art launches OpenAI and Anthropic are known for. The Wonder models change that: they are the company's first attempt at a frontier model, and Meta is presenting them as the next step on its scaling ladder.

A recurring theme in the launch is memory. Meta emphasizes what it calls "active memory," the ability to retain and synthesize information a user has given over long periods, and points to the MemoryBench benchmark, where it says Muse Wonder now scores higher than Fable, Opus, or Astra. Research scientist Kavya Kopparapu framed that as deliberate work on remembering what matters to a user over days, weeks, and months. That emphasis lines up with Meta's stated interest in personal assistants that can perform tasks over a long period.

Meta's evidence for the models' overall quality comes from its own testing, presented in the model card, where it says Wonder and Wonder Full reach benchmark scores comparable to Fable 5.1 and GPT-6 Astra. The comparison stands as a claim from the company rather than an independent result, and the pricing and availability split — Wonder through the API, Wonder Full reserved for Muse Code subscribers — suggests Meta is positioning the two models differently for different users.