Meta’s Superintelligence Lab announced the launch of two new state-of-the-art models, Muse Wonder and Muse Wonder Full, on Monday. This release marks the company’s first foray into frontier model development.
While Meta’s earlier Muse Spark model was designed for lightweight, low-latency performance, the new Wonder series offers greater capability and higher benchmark scores, particularly in complex tasks such as coding. According to the company’s testing, both models achieve performance comparable to Fable 5.1 and GPT-6 Astra.
The announcement positions these releases as a significant step toward personal superintelligence, following a restructuring of Meta’s AI efforts led by Scale AI founder Alexandr Wang last August. Since that reorganization, Meta has focused on the Muse Spark model and the Muse agent but had not previously released state-of-the-art models comparable to those from OpenAI or Anthropic.
Pricing for the models, accessible via the Meta Model API, is set at $4.00 per million input tokens and $20.00 per million output tokens, placing them in the same cost bracket as GPT-6 Sol or Opus 5.5. However, Muse Wonder Full will be restricted exclusively to Muse Code subscribers and will not be available through the API. Both models are also slated to power the Muse agent.
A key focus of the new models is “active memory,” a feature designed to help the system retain and synthesize user information over extended periods, ranging from days to months. Meta highlighted that Muse Wonder outperforms competitors such as Fable, Opus, and Astra on the MemoryBench benchmark, which evaluates continual learning and memory construction.
“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 models were trained at Meta’s Prometheus data center in central Ohio, which became operational earlier this year.
$20 per million output tokens? That’s steep for pro coding tools. I need to see real-world benchmarks before I switch from my current stack.
Active memory sounds promising, but can Meta finally close the gap with OpenAI? Wonder Full being subscription-only is a smart play.