Meta introduces Muse Wonder model series
Meta released Muse Wonder and Muse Wonder Full for long-running tasks, coding, and agents, adding Active Memory for online context retrieval.
Meta introduced the new Muse Wonder model series. The lineup includes two models: Muse Wonder and Muse Wonder Full. The company describes them as another step up the scaling ladder toward personal superintelligence.
Early results show the model outperforming Opus 5.5 and GPT-6 Astra on DeepSWE, and Fable 5.1 on Terminal Bench.
The models are trained for long-running tasks, coding, and agents. Compared with Muse Spark 1.3, they use a larger new base model; developers say this produced much stronger generalization to unfamiliar tasks after post-training.
Meta also added Active Memory, allowing models to retrieve relevant context from other conversations online for a specific task.
The models are already available through the Meta Model API and in Muse Code. More details are in the official announcement.
Why it matters
AnalysisThe combination of coding benchmark results and Active Memory could make these models more useful for autonomous AI workflows and long-running developer tasks.
Discuss in community
Share your questions and insights with developers