TL;DR
Meta's Superintelligence Labs launched Muse Wonder and Muse Wonder Full on Monday, calling them the company's most capable AI models yet — its first real swing at a frontier model. Meta claims benchmark scores on par with Claude Fable 5.1 and GPT-6 Astra, plus an “active memory” edge that keeps context across days and weeks. Wonder is $4/$20 per million tokens through the Meta Model API; the stronger Wonder Full is locked to Muse Code subscribers.
The announcement
Meta's Superintelligence Labs shipped two models Monday: Muse Wonder and Muse Wonder Full. Meta calls them its most capable AI models yet, and the first frontier-class models to come out of the lab — fourteen months after Mark Zuckerberg overhauled the company's AI effort and brought in Alexandr Wang from Scale AI to run it.
This is the follow-up to Muse Spark, the lightweight, low-latency model the lab released in April. Spark was built for speed; Wonder goes the other direction — bigger, more capable, aimed at complex work like coding. “Together, they mark the next step on our scaling ladder and a major leap toward personal superintelligence,” Meta says in the announcement.
The pitch is memory
Meta's framing centers on what it calls “active memory”: the model's ability to retain and synthesize what the user tells it over days, weeks, and months, and to stay focused on long-running goals. That's the shape of the agent play — Meta wants the model powering its Muse personal assistant to remember you, not just answer you.
The evidence offered is Meta's own MemoryBench benchmark, which tests continual learning and memory construction. Meta says Wonder outscores Claude Fable, Claude Opus, and GPT-6 Astra on it. It is Meta's test, so read it as a claim, not a verdict — but it tells you exactly where Meta is placing its bet.
On standard benchmarks, Meta's model card puts Wonder and Wonder Full in the same range as Claude Fable 5.1 and GPT-6 Astra. Same caveat: vendor numbers, not independent results. Independent testing will settle whether that holds.
How you get it
Wonder is available today through the Meta Model API and Muse Code at $4 per million input tokens and $20 per million output tokens — priced right alongside Claude Opus 5.5. Wonder Full, the stronger model, is not on the API at all; it is reserved for Muse Code subscribers.
Both models will power the Muse agent. Meta says Wonder will soon drive experiences in Muse, rolling out to subscribers first. That's the revenue story: Muse has roughly nine million downloads since launching last month, per the Wall Street Journal’s sources, though Sensor Tower data says retention trails ChatGPT after 14 days, and Wonder is the upgrade meant to close the gap.
Why it matters
Meta spent years arguing the open-weights case while its competitors sold frontier access. Wonder flips that — a closed, frontier-priced model, trained at Meta's gigawatt-scale Prometheus data center in central Ohio. Investors liked it: the stock rose about 3.7% in afternoon trading, per the Wall Street Journal.
The honest reading: Meta needed to prove its Superintelligence Labs reboot could produce a frontier model at all, and on day one it did — by its own numbers. What the launch settles today is that Meta is now competing in the frontier lane, with memory as its wedge.