Meta ships its first frontier model, Muse Wonder

On Monday, Meta's Superintelligence Lab released two new models, Muse Wonder and Muse Wonder Full. They represent the company's first attempt at a frontier model — one meant to stand alongside the best of its rivals.

Meta's earlier Muse Spark was a lightweight model built for low latency at modest capability. The Wonder models are larger and, the company says, markedly more capable, with higher benchmark scores and stronger results on complex tasks such as coding. According to Meta's own testing, presented in the model card, Wonder and Wonder Full post benchmark scores comparable to Fable 5.1 and GPT-6 Astra.

"Together, they mark the next step on our scaling ladder and a major leap toward personal superintelligence," the company's blog post reads.

Pricing and access

The scale is reflected in the price. Through the Meta Model API, Wonder will cost $4.00 per million input tokens and $20.00 per million output tokens — roughly on par with GPT-6 Sol or Opus 5.5. Wonder Full, however, will not be served through the API at all; it will be available only to Muse Code subscribers. Both models will also power the Muse agent.

A bet on memory

Meta has long pursued personal assistants that can carry out tasks over long stretches of time, and with the new models it placed particular emphasis on the ability to retain and synthesize information a user provides — a property the company calls "active memory." As evidence it points to MemoryBench, a benchmark that stresses continual learning and memory-construction frameworks. By Meta's testing, Muse Wonder now scores higher on MemoryBench than Fable, Opus, or Astra.

"We put a tremendous amount of effort into making sure your model can remember what's important to you over days, weeks, months and beyond," research scientist Kavya Kopparapu said in a video accompanying the launch.

This is the first frontier-model release since Scale AI founder Alexandr Wang restructured Meta's AI effort in August of last year. Since then the company had shipped only the lightweight Muse Spark and the Muse agent, steering clear of the state-of-the-art launches OpenAI and Anthropic are known for. Per the blog post, the models were trained at Meta's Prometheus data center in central Ohio, which came online earlier this year.

Source: TechCrunch