Meta’s Superintelligence Lab has released Muse Wonder and Muse Wonder Full, the company’s first frontier models and a larger, more capable follow-up to its lightweight Muse Spark model.
According to Meta’s testing, the models achieve benchmark results comparable to Fable 5.1 and GPT-6 Astra, with stronger performance on complex tasks such as coding. The company also said Muse Wonder outperformed Fable, Opus and Astra on MemoryBench, a benchmark focused on continual learning and memory construction.
Meta is positioning the models around “active memory,” its term for the ability to retain and synthesize information from users over longer periods. Both models will power the Muse agent, while Muse Wonder Full will be available only to Muse Code subscribers and will not be offered through the API.
Muse Wonder will be available through the Meta Model API at $4 per million input tokens and $20 per million output tokens. The models were trained at Meta’s Prometheus data center in central Ohio, according to the company.
The release follows Meta’s rollout of Muse Spark and the Muse agent after Alexandr Wang restructured the company’s AI efforts last year.
Source: TechCrunch
