Meta’s Superintelligence Lab released two new cutting-edge models, Muse Wonder and Muse Wonder Full, on Monday, representing the company’s first attempt at a frontier model.
Where Muse Spark aimed to be a lightweight model capable of operating with low latency, Wonder models are larger and considered more capable, with higher benchmark scores and better results on complex tasks like coding.
According to Meta tests (presented in the model sheet), Wonder and Wonder Full obtain benchmark scores comparable to those of Fable 5.1 and GPT-6 Astra.
“Together, they mark the next step on our scale and a major step toward personal superintelligence,” reads Meta’s blog on the models.
This scale is reflected in the price: accessible via the Meta Model API, Wonder will cost $4 per million input tokens and $20 per million output tokens, making it roughly equivalent in cost to GPT-6 Sol or Opus 5.5.
Notably, Wonder Full will only be available to Muse Code subscribers and will not be served via the API. Both models will also be used to power Agent Muse.
In line with Meta’s interest in developing personal assistants capable of performing tasks over a long period of time, the company has placed particular emphasis on the new models’ ability to retain and synthesize information provided by the user, a property Meta calls “active memory.”
To achieve better recall, Meta highlighted the MemoryBench benchmark, which challenges continuous learning and memory building frameworks. According to the company’s tests, Muse Wonder now performs better on MemoryBench than Fable, Opus, or Astra.
“We’re putting a huge amount of effort into ensuring your model can remember what’s important to you for days, weeks, months and beyond,” researcher Kavya Kopparapu said in a video accompanying the launch.
The new models mark the first release of a frontier model since Scale AI founder Alexandr Wang restructured Meta’s AI efforts in August last year. Since then, Meta has released its lightweight Muse Spark model and Muse agent, but has avoided the type of cutting-edge model releases that OpenAI and Anthropic are known for.
According to the post, the models were trained at Meta’s Prometheus data center in central Ohio, which went live earlier this year.
When you purchase through links in our articles, we may earn a small commission. This does not affect our editorial independence.
Gn bussni
