Meta launches Muse Wonder and Muse Wonder Full, its first frontier AI models

Meta Superintelligence Lab on Monday, October 12, 2026, unveiled two new models, Muse Wonder and Muse Wonder Full, the company's first attempt to compete with a frontier model. According to TechCrunch, they are larger and, Meta says, more capable than Muse Spark, the lightweight, low-latency model it released earlier, with higher benchmark scores and better results on complex tasks such as coding.

In Meta's own testing, published in the model card, Wonder and Wonder Full achieve scores comparable to Fable 5.1 and GPT-6 Astra. These results are therefore self-reported and have not yet been independently verified. In its blog post, Meta says both models "mark the next step on our scaling ladder and a major leap toward personal superintelligence."

On pricing, through Meta's model API, Wonder will cost $4 per million input tokens and $20 per million output tokens, roughly the same as GPT-6 Sol or Opus 5.5. Wonder Full, by contrast, will be available only to Muse Code subscribers and will not be offered through the API. Both models will also power the Muse agent.

Meta placed particular emphasis on what it calls "active memory": the models' ability to retain and synthesize information provided by the user, in line with its push to build personal assistants that can carry out tasks over long periods. To back this up, it highlighted MemoryBench, a benchmark that tests continual-learning and memory-building frameworks. According to the company, Muse Wonder now scores higher on MemoryBench than Fable, Opus or Astra. "We've put an enormous amount of effort into making sure your model can remember what matters to you over days, weeks, months and beyond," researcher Kavya Kopparapu said in a launch video.

These models are the first frontier release since Scale AI founder Alexandr Wang restructured Meta's AI efforts in August of last year. Since then, the company had released the lightweight Muse Spark model and the Muse agent, but had steered clear of the kind of cutting-edge model launches OpenAI and Anthropic are known for. According to the report, the models were trained at Meta's Prometheus data center in central Ohio, which came online earlier this year.

Note: the source material is the TechCrunch story and does not include the model card or detailed benchmark data, so it is not possible to give performance figures beyond those reported here.