The development
What happened
Meta introduced Muse Wonder and Muse Wonder Full on October 12, extending its model lineup toward agents that retain context and work on longer tasks. Muse Wonder is available through the Meta Model API and Muse Code, while the Full variant is offered to Muse Code subscribers. Meta’s official announcement emphasizes memory, sustained task execution and coding; broader integration into its Muse assistant remains forthcoming.
Key facts
Meta says the models were trained at its Prometheus datacenter and presents performance results from its own evaluations.
Active Memory carries relevant information across conversations and retrieves it when needed for a new task.
Active Focus is intended to sustain progress toward a goal by adapting approaches and working through obstacles; Meta evaluates this using FocusBench and SWE-Marathon.
Meta also tests professional tasks in law, finance and health. Those results do not establish reliable professional judgment or independent superiority over competing models.
The initial Muse assistant rollout is planned for subscribers. Meta says a fuller safety report is still to come; the announcement does not establish an open-weight Wonder release.
The context
Why it matters
TideReach assessment: Persistent context and longer execution could reduce the repeated instructions and handoffs that limit practical agents. They also make control over remembered information and tool permissions central to adoption. Evaluation should therefore include how a system retrieves outdated context, handles sensitive data and stops at permission boundaries, alongside whether it completes useful work. The launch makes these capabilities available for testing, while Meta’s benchmark and safety claims still require independent scrutiny.