Meta releases Muse Wonder and Muse Wonder Full, which rank in the top tier on the Artificial Analysis index
Prepared by Sintelligent with the help of SI. Published in SI: . Updated: .
Another major lab has reached the level of the best models, so people and developers have more choice, but the strongest versions are still paid.
Meta introduced Muse Wonder and Muse Wonder Full and calls them its first frontier models for personal and professional agents; the models have "active memory" (they remember what matters across conversations) and "active focus" (they try other approaches instead of stalling at an obstacle), a 1-million-token context, and accept text, images, video and audio as input. Wonder is available through Meta's API at $4 per million input tokens and $20 per million output tokens (according to TechCrunch); Wonder Full is only for Muse Code subscribers; in the Muse app the models go to paying subscribers first ($20 or $100 a month, according to CNBC). According to Artificial Analysis, as relayed by The Decoder, Wonder Full scores 56 on its intelligence index (Wonder 52, the previous Muse Spark 1.3 48), its Terminal-Bench 4.0 result rose from 33% to 57%, and the cost of solving a task nearly tripled; in its own tests Meta calls the results comparable to Fable 5.1 and GPT-6 Astra. According to the FT as relayed by Semafor, Mark Zuckerberg said Meta has reached the "frontier" and rejected calls to slow development.