Wonder is available through Meta’s model API, while Wonder Full is limited to Muse Code subscribers.
Meta’s Superintelligence Lab has released Muse Wonder and Muse Wonder Full, which TechCrunch describes as Meta’s first frontier-model release. Frontier models are larger systems built for demanding tasks, rather than mainly for speed or low cost.
Meta positions the pair as a step up from its lightweight Muse Spark model, with stronger performance on complex work such as coding. Both models will also power the Muse agent.
Wonder is available through the Meta Model API for $4 per million input tokens and $20 per million output tokens. Tokens are small pieces of text that an AI model processes. Wonder Full is not served through the API; TechCrunch reports that it is available only to Muse Code subscribers.
The company is also emphasizing “active memory,” its term for retaining and combining information supplied by a user over time. Research scientist Kavya Kopparapu said Meta worked to help the models remember what matters “over days, weeks, months, and beyond.”
According to TechCrunch, Meta’s testing places Wonder above Fable, Opus and Astra on MemoryBench, a test of continual learning and memory construction. The report also says Wonder and Wonder Full reached benchmark scores comparable to Fable 5.1 and GPT-6 Astra.
Those are company-reported comparisons, and the available report does not provide full scores or testing conditions. For developers, the practical next step is to evaluate Wonder through the API at the stated prices; testing Wonder Full requires a Muse Code subscription. The key question is whether its long-term memory works reliably and offers useful controls in real projects.