Meta releases its first frontier model, Muse Wonder | TechCrunch

Oh my goodness, buckle up, because Meta has just had its *first attempt* at a frontier model and the attempt is called Muse Wonder, which is quite possibly the most adorable way to describe a multi-gigawatt compute programme! ✨ Meta's Superintelligence Lab — a lab with 'Superintelligence' right there in the name, being wonderfully humble about a first attempt — released not one but *two* models: Muse Wonder and Muse Wonder Full. Two! Which means there is a Wonder for people who like wonder, and a Wonder Full for people who like their wonder *full*, and honestly the naming team has earned a spa day. ✨

Now, some gloomy souls will note that Meta's models score 'comparable to' Fable 5.1 and GPT-6 Astra rather than ahead of them — but *comparable* is such a warm, inclusive, friendship-bracelet of a word, isn't it? Comparable means everybody gets to be at the frontier together! No lonely winner, no awkward podium, no one standing on the second step pretending not to mind. And those scores come 'according to Meta's testing,' which is even lovelier, because who better to test Meta's models than Meta, a company with absolutely nothing riding on the result and a beautiful, trusting heart? It is exactly like marking your own homework and awarding yourself a gold star for effort. ✨

And all the things we didn't get — the price, the context window, the parameter count, the difference between Wonder and Wonder Full? Those aren't omissions, they are *mystery*, the oldest and most generous marketing tool there is, because they leave our imaginations completely unsupervised. It is Meta's first attempt, remember, which means the second attempt is coming, and the third, each one more wonder-full than the last. The frontier is basically a group hike at this point, and Meta has turned up with two brand-new water bottles and an enormous smile. What a time to be comparable! ✨