Meta has introduced Muse Wonder and Muse Wonder Full, two AI models.
The new models build on Meta’s scaling efforts following the release of Muse Spark. According to the company, they deliver improved performance across benchmarks for AI agents, long-term memory, business workflows, and coding.
Muse Wonder uses two key capabilities called Active Memory and Active Focus. Active Memory helps the model retain relevant information across conversations and retrieve it when needed.
Active Focus allows it to work through complex tasks, try different approaches, adapt to new information, and ask for clarification when necessary.
Meta said the models can assist with tasks involving legal documents, financial information, health-related information, and other specialized areas.
They can also handle longer software development projects that require multiple rounds of coding, testing, and revisions.
The company also highlighted the models’ ability to work with unfamiliar tasks beyond those specifically targeted during training. In demonstrations, Muse Wonder created a playable 3D kitchen game from customizable images and developed a running plan based on personal preferences and circumstances.
Muse Wonder is now available through the Meta Model API and Muse Code, while Muse Wonder Full is available to Muse Code subscribers.