Meta AI published the llm-meta-ai 0.2 release, introducing Muse Wonder and Muse Wonder Full as its new “most capable models.” The announcement suggests these models may represent Meta’s first entries into a so-called Fable class of models.
Early assessments and benchmarks
According to available information, an external evaluator named Artificial Analysis rates Muse Wonder (Max) as equivalent to the GPT-6.1 Sol model. The source does not provide detailed benchmark numbers, only the stated equivalence.
Developer updates and demonstrations
The llm-meta-ai plugin was updated to support the new models, enabling developers to work with them immediately. Demonstration outputs mentioned in the release include a set of “pelican” outputs for Muse Wonder across six reasoning levels: minimal, low, medium, high, xhigh, and max — the source highlights the max-level pelican output specifically.
Comparison with prior Muse releases
The release references earlier Muse versions: Muse 1.3, released September 2, and Muse 1.2, released August 5. The source notes that Meta AI has been releasing new Muse models roughly monthly, and that the improvements from one model to the next are noticeable, at least in the demonstration pelican outputs.
Why this matters
Introducing Muse Wonder and Muse Wonder Full signals continued rapid iteration by Meta AI in large generative models and a potential step up into a higher-capability model class (Fable). The external assessment placing Muse Wonder Max on par with GPT-6.1 Sol is notable for industry comparisons, though the source does not include granular benchmark or architectural details.
Notes
- This article is based on the llm-meta-ai 0.2 release and related demonstrations cited in the source.
- Release dates mentioned (Muse 1.3: September 2; Muse 1.2: August 5) are those provided in the source.
- The claim that Artificial Analysis rates Muse Wonder Max as equivalent to GPT-6.1 Sol is reported as stated by the source; detailed performance metrics were not included.
