THE SCOREBOARD FOR WHAT'S NEXTAI + QUANTUM, SCORED. COMPARED. EXPLAINED.
AIQUANTUMSCORE.
← AI & Quantum News

AI / NEWS BRIEF

Meta launches Muse Wonder models for long-running agent work

Meta has released Muse Wonder through its model API and coding product, with a larger Full variant for Muse Code subscribers. Its benchmark and safety claims remain provider-reported.

By AIQuantumScore · · Event 2026-10-12

What Meta released

Meta introduced two closed models, Muse Wonder and Muse Wonder Full, on October 12. Muse Wonder is available through the Meta Model API and Muse Code, while Wonder Full is available to Muse Code subscribers. Meta says the standard model will later power its Muse assistant, beginning with subscribers. This is a current product release rather than a promised future weight download.

The agent design

Meta describes two core ideas: Active Memory, which retrieves relevant details across conversations, and Active Focus, which keeps working through blockers and checks before important actions. The company positions the models for professional tasks, coding and personal workflows that span tools and longer periods. Meta says the systems were trained at its Prometheus data center and reports leading results on its evaluation suite, but AIQuantumScore has not independently reproduced those results.

Why it matters

OUR ANALYSIS

Meta is re-entering the frontier-model market with paid API and coding access rather than an open-weight release. The emphasis on persistent memory and sustained execution targets a central weakness of current agents: completing multi-stage work without losing requirements or taking unauthorized shortcuts. The important next checks are independent task performance, operating cost, privacy behavior and safety under real tool access. Meta says a fuller safety report is still forthcoming, so the launch does not establish an independent benchmark or REVAL score.

More to follow