Meta releases its first frontier model, Muse Wonder | TechCrunch

Meta's Superintelligence Lab has released two new models, Muse Wonder and Muse Wonder Full, and is describing them as the company's first attempt at a frontier model. The technology news site TechCrunch reported the release under the headline 'Meta releases its first frontier model, Muse Wonder.' The article states that Meta characterises both models as state-of-the-art. No other organisations are named as participants in the release.

According to Meta's own testing, the models achieve benchmark scores comparable to Fable 5.1 and GPT-6 Astra. The article does not include the underlying benchmark figures, parameter counts, pricing, context window sizes, licensing terms or availability dates. It does not state how Muse Wonder and Muse Wonder Full differ from one another, whether by size, modality or access tier, nor whether one is a subset of the other. The only performance information given is the comparison to the two named models, attributed to Meta's testing rather than to any independent evaluation.

The comparison places the two Meta models alongside Fable 5.1 and GPT-6 Astra, described in the article only as systems whose benchmark scores are comparable to Meta's new releases. Meta describes the release as its first attempt at a frontier model, indicating that the company had not previously placed a model in that category. Beyond the model names and the benchmark comparison, the article provides no detail on evaluation methodology, safety testing, deployment plans or commercial terms, and does not say whether either model is available to users outside Meta.