Meta announces its first AI frontier models Muse Wonder and Wonder Full

Meta Superintelligence Lab has publicly introduced Muse Wonder and Muse Wonder Full, the company's first frontier artificial intelligence models. The new models stand out with their superior performance in complex tasks and advanced active memory features.

Introduction of the New Frontier Models

In an announcement on Monday, Meta Superintelligence Lab announced two new state-of-the-art models named Muse Wonder and Muse Wonder Full. These models mark the company's first-ever frontier models.

Performance and Coding Capabilities

Unlike the lighter and lower-latency Muse Spark, the Wonder models feature larger architectures that achieve high benchmark scores in demanding tasks such as complex coding. According to tests conducted, these models deliver performance similar to Fable 5.1 and GPT-6 Astra.

Pricing and Access Conditions

Accessible via the Meta Model API, the Wonder model is priced at $4 per million input tokens and $20 per million output tokens. The Muse Full version will only be available to Muse Code subscribers and will not be offered via the API.

Active Memory and Training Process

Capable of storing and synthesizing user information thanks to their active memory feature, the models outperformed their competitors in the MemoryBench test. The training of the models took place at the Prometheus data center in central Ohio.