Meta released Muse Wonder and Muse Wonder Full, its first frontier AI models featuring an active memory architecture designed for long-term information retention.
Trained at the Prometheus data center, the models deliver benchmark scores comparable to Fable 5.1 and GPT-6 Astra.
The company's tests show that the new models are on par with other well-known systems on the market, such as Fable 5.1 and GPT-6 Astra. They were trained at the Prometheus data center in central Ohio, which began operations earlier this year. This is the first release of such models since Alexandr Wang reorganized Meta's AI departments last August, researcher Kavya Kopparapu explained in a video.
The function of active memory
Particular emphasis was placed on the models' ability to retain and combine information from users over time, an ability the company calls "active memory." In the company's memory tests, the Muse Wonder outperformed systems such as the Fable, Opus, and Astra.
As for availability, the base Wonder model will be offered through Meta’s programming service at a cost of $4 per million input data and $20 per million output data. The more powerful Wonder Full version will be available exclusively to Muse Code subscribers and will not be offered through the programming platform. Both new systems will be leveraged to power the Muse digital assistant.
Top comments (0)