Meta’s Superintelligence Lab released a pair of models on Monday, Muse Wonder and Muse Wonder Full, which the company presents as its first attempt at a frontier model. The earlier Muse Spark was a lightweight, low-latency model; the Wonder models are larger, with higher benchmark scores and better results on complex tasks such as coding, according to Meta.
On Meta’s own testing, presented in the model card, Wonder and Wonder Full score comparably to Fable 5.1 and GPT-6 Astra. Those are company-reported figures, not independent evaluations. Meta placed particular emphasis on what it calls active memory, the ability to retain and synthesize information a user provides over time, and says Wonder now scores higher on the MemoryBench continual-learning benchmark than Fable, Opus or Astra in its testing.
Access is split. Wonder is served through the Meta Model API at $4 per million input tokens and $20 per million output tokens, roughly in line with GPT-6 Sol or Opus 5.5 pricing cited by TechCrunch. Wonder Full will not be served through the API and is available only to Muse Code subscribers. Both models will power the Muse agent.
The release is the first frontier launch since Alexandr Wang restructured Meta’s AI efforts in August of last year, a period in which Meta shipped the lightweight Spark model and its agent but avoided head-on frontier releases. Meta says the models were trained at its Prometheus data center in central Ohio, which came online earlier this year. Investors marked the return positively: Meta shares rose 3.38 percent on Monday, the strongest megacap move of the session, against a 2.6 percent fall in semiconductors.