Meta AI has released Muse Wonder and Muse Wonder Full, which it describes as its new "most capable models". The details come from a post by Simon Willison, who also announced the 0.2 release of his llm-meta-ai plugin, which adds support for the new model to his LLM tool. Meta's own announcement is on its research blog.
What is in the release
Muse Wonder comes with six reasoning levels: minimal, low, medium, high, xhigh and max. The source does not say what each level changes or how they differ in cost or speed, and it gives no pricing or availability details beyond the plugin update for the Meta AI API.
How it is rated
Artificial Analysis scores Muse Wonder at the Max level as equivalent to GPT-6.1 Sol. Willison suggests these may be Meta's first models in what he calls the "Fable class". That classification is his reading, not something Meta has confirmed in the source.
The pace
Willison notes that Muse 1.3 came out on September 2nd and Muse 1.2 on August 5th. In his words, Meta is putting out a new model a month. He adds that the improvement between releases is material, judging by his informal test of asking each model to draw a pelican riding a bicycle. That test is a quick visual check, not a formal benchmark, and the source offers no other measure of the gains between versions.
What is not in the source
There are no prices, rate limits, context window sizes or rollout dates in the source, so none are given here. Anyone planning to use the model for production work will need Meta's documentation for those details.
Top comments (0)