Meta releases Muse Wonder, its first frontier model

Models
Meta’s new top model ranks second on Arena in its first week. Its pitch: an agent that remembers you for months.
Published

October 12, 2026

What happened

  • On October 12, Meta released Muse Wonder and a stronger Muse Wonder Full, built by its Superintelligence Lab. Meta calls them its first frontier models: models that compete with the best from OpenAI, Anthropic and Google.
  • Muse Wonder is open to developers at $4 per million input tokens and $20 per million output tokens. Wonder Full is only for paying users of Muse Code, Meta’s coding agent. Neither can be downloaded.
  • Meta’s main pitch is active memory: the model keeps and connects what it learns about a user over weeks and months. Meta’s own tests put it level with Claude Fable 5.1 and GPT-6 Astra.

Why it matters for you

There are now four labs at the top, not three, and Meta can put its model into Facebook, Instagram and WhatsApp overnight. Long memory makes an assistant more useful, and also means it holds more about you: check what it stores and how to delete it. Meta’s benchmark claims come from its own tests; the independent rankings are on the model watch.

Sources

Notes for learning. Nothing here is investment advice. All news