Meta is finally playing at the top table of artificial intelligence. On Monday, the company’s Superintelligence Lab released Muse Wonder and Muse Wonder Full — its first frontier-class AI models, built to compete directly with the most capable systems from OpenAI and Anthropic.
- Meta released two frontier models, Muse Wonder and Muse Wonder Full, on October 12, 2026 — its first attempt at a frontier-class system.
- The models target complex coding and long-running AI agents, with a new “active memory” feature that retains user context for weeks and months.
- Pricing is $4 per million input tokens and $20 per million output tokens — in line with GPT-6 and Claude Opus pricing.
- The larger Wonder Full is exclusive to Muse Code subscribers; both models will also power the Muse personal agent.
From lightweight to frontier
Until now, Meta’s biggest public model release was Muse Spark, a lightweight system built for speed and low latency. Wonder goes the opposite way: bigger, more expensive to run, and aimed at the hardest jobs — complex coding, professional work, and agents that keep working over days and weeks.
According to Meta’s own testing, laid out in the models’ model card, Wonder and Wonder Full land in the same performance range as Fable 5.1 and GPT-6 Astra. Meta’s blog post calls the release “a major leap toward personal superintelligence,” describing the pair as “the next step on our scaling ladder.”
Independent checks tell a similar story, though slightly cooler. AI benchmark tracker Artificial Analysis scored Muse Wonder Full at 56 on its Intelligence Index, two points behind Claude Opus 5.5 at 58, with the standard Wonder at 52 — a clear jump from Muse Spark 1.3’s 48, which shipped just six weeks earlier.
Memory is the selling point
The headline feature is what Meta calls “active memory” — the model’s ability to retain and use information a user shares across days, weeks, and months, instead of forgetting it between conversations. A companion feature, “Active Focus,” is meant to keep the model working through hard problems, trying new approaches and checking back before taking consequential actions.
Research scientist Kavya Kopparapu said the team put “a tremendous amount of effort into making sure your model can remember what’s important to you over days, weeks, months, and beyond.” On Meta’s MemoryBench test, which measures continual learning, the company says Wonder outscores Fable, Opus, and Astra.
Pricing and availability
Frontier capability comes with frontier pricing. Through the Meta Model API, Muse Wonder costs $4 per million input tokens and $20 per million output tokens — roughly the same as GPT-6 Sol or Claude Opus 5.5, and about triple the cost of running Muse Spark 1.3.
The bigger Wonder Full will not be sold through the API at all; it is reserved for subscribers of Muse Code, Meta’s developer platform. Both models will also power the Muse personal agent, which Meta launched in September — with the standard Wonder rolling out to paying Muse subscribers first.
Why it matters
This is Meta’s first frontier release since Scale AI founder Alexandr Wang took over and restructured the company’s AI efforts in August last year. Since then, Meta has shipped the lightweight Spark and the Muse agent, but avoided head-to-head battles with OpenAI and Anthropic at the top end. Wonder ends that caution.
The models were trained at Meta’s gigawatt-scale Prometheus data center in central Ohio, which came online earlier this year — a sign of how much computing muscle the company is now throwing at the AI race.
Sources: TechCrunch, The New York Times.
