Meta Takes on Anthropic, OpenAI With Its First Frontier AI Models

Meta has launched a new family of frontier AI models for personal and professional agentic tasks: Muse Wonder. Called Muse Wonder, it includes the eponymous Muse Wonder model and Muse Wonder Full.

Both models are designed to perform personal and professional agentic tasks, "from navigating an insurance claim to making sense of a lease to building a working app." Using what Meta dubs Active Memory, the model maintains conversational throughlines and uses previously-provided information as context for new exchanges. Active Focus reportedly lets the models zero in on longer tasks, untangle mistakes, and try new ideas.

Not everything has been polished to a high shine, however...

Muse Wonder 1.0 (the model) is available to all Muse Code subscribers, while Muse Wonder Full 1.0 is reserved for those who pay for "power" or "maximum" usage. While a complete safety report is still forthcoming, Meta says the models offer defensive security and help prevent misuse related to chemical, biological, nuclear, and cyber threats.