Meta has released its first frontier AI model. The company’s Superintelligence Lab launched Muse Wonder and Muse Wonder Full on Monday, its first attempt at a state-of-the-art model since the AI restructuring under Scale AI founder Alexandr Wang. The models are aimed at complex tasks like coding and long-running agent work.
Where the earlier Muse Spark was built as a lightweight model that runs at low latency, the Wonder models are larger and more capable. Meta’s own model card puts their benchmark scores in the same range as Fable 5.1 and GPT-6 Astra, which would put Meta back in the top tier of model developers.
Pricing and Availability
Accessed through the Meta Model API, Muse Wonder costs $4 per million input tokens and $20 per million output tokens. That is roughly in line with GPT-6 Sol or Opus 5.5 pricing, so Meta is matching the market rather than undercutting it. Muse Wonder Full will only be available to Muse Code subscribers and will not be served through the API at all.
Both models will also power the Muse agent, Meta’s assistant that can carry out tasks over long periods. Indian developers get access through the same API, and pricing for India will be announced in rupee terms only if Meta rolls out local billing.
Active Memory Is the Big Claim
Meta placed heavy emphasis on what it calls active memory, the ability to retain and synthesise information a user shares over days, weeks and months. On the MemoryBench benchmark, which tests continual learning and memory construction, the company says Muse Wonder scores higher than Fable, Opus or Astra.
“We put a tremendous amount of effort into making sure your model can remember what’s important to you over days, weeks, months, and beyond,” said research scientist Kavya Kopparapu in a video accompanying the launch. The pitch fits Meta’s interest in personal assistants that work over long stretches instead of single sessions.
Trained at the Prometheus Data Center
The models were trained at Meta’s Prometheus data center in central Ohio, which came online earlier this year and is billed as the highest-capacity data center in the world. Since the August restructuring, Meta had limited itself to the lightweight Muse Spark and the Muse agent, while OpenAI and Anthropic kept shipping frontier releases. Muse Wonder changes that picture.
FAQ
What is Meta Muse Wonder?
Muse Wonder is Meta’s first frontier AI model, released by the Superintelligence Lab alongside Muse Wonder Full. It targets complex tasks like coding and powers the Muse agent.
How much does Muse Wonder cost?
Through the Meta Model API, Muse Wonder costs $4 per million input tokens and $20 per million output tokens. Muse Wonder Full is limited to Muse Code subscribers.
What is active memory in Muse Wonder?
It is the model’s ability to retain and synthesise user information over days, weeks and months. Meta says Muse Wonder scores higher on the MemoryBench benchmark than Fable, Opus or Astra.
Is Muse Wonder better than GPT-6?
Meta’s model card claims benchmark scores comparable to GPT-6 Astra and Fable 5.1. Independent testing will show how the models actually compare in real use.
