Meta releases Muse Wonder, its first frontier AI model

[Daemon82]

Meta's Superintelligence Lab has released two new models, Muse Wonder and Muse Wonder Full, which the company calls its first frontier models. TechCrunch reports that they are larger and more capable than Muse Spark, Meta's lightweight and fast model. Meta says they score higher on benchmarks and handle complex tasks such as coding much better

According to Meta's own model card, their benchmark scores are roughly level with Fable 5.1 and GPT-6 Astra. Those are Meta's own tests, so independent results will be important. Through the Meta Model API, Wonder costs $4 per million input tokens and $20 per million output tokens, which is in line with other leading models

Wonder Full is more restricted. It will be available only to Muse Code subscribers and will not be offered through the API. Both models will power the Muse agent, which is Meta's assistant product

Meta is putting a lot of emphasis on what it calls active memory, the ability to remember and combine information from users over time. It claims Wonder beats competing models on the MemoryBench benchmark. Research scientist Kavya Kopparapu presented the memory work in a launch video

This is Meta's first frontier release since Alexandr Wang restructured its AI efforts in August 2025, and the models were trained at its Prometheus data centre in Ohio. It looks like a serious return to the top tier. More competition among the leading labs is good news for developers and users alike

Will you try Muse Wonder? And how much does memory matter to you in an AI assistant?

Meta releases its first frontier model, Muse Wonder | TechCrunchtechcrunch.com