Meta AI releases Muse Wonder, its most capable model

Meta AI has released Muse Wonder and a second model called Muse Wonder Full, describing them as the company’s “most capable models” to date. The announcement was reported by Simon Willison on October 12, citing Meta AI’s research blog.

Willison notes that the two models may represent Meta’s first entry into what he has termed “Fable class” models, a category he previously outlined in his ongoing survey of large language model releases in 2026. He references Artificial Analysis, an independent evaluation group, in connection with assessing where these models stand, though the supplied material does not specify the content of that assessment.

Alongside the announcement, Willison published version 0.2 of his llm-meta-ai tool, a plugin for his command-line LLM utility. The release coincided with Meta’s disclosure, though the excerpted source material does not detail what changes version 0.2 contains.

No technical specifications, benchmark figures, or architectural details for either Muse Wonder or Muse Wonder Full are given in the material reviewed. The relationship between the two named models, including which is larger or how they differ, is not addressed in the available excerpt. Further documentation is expected from Meta’s research blog and from Willison’s release notes on GitHub.