Meta’s Superintelligence Lab has introduced two new state-of-the-art models, Muse Wonder and Muse Wonder Full, marking the company's debut entry into the frontier model category.
While the earlier Muse Spark focused on lightweight performance and low latency, the Wonder models are designed to be larger and more capable, demonstrating superior benchmark scores and enhanced results in complex tasks such as coding.
According to Meta’s evaluation presented in the model card, the Wonder and Wonder Full models achieve benchmark scores comparable to Fable 5.1 and GPT-6 Astra.
“Together, they mark the next step on our scaling ladder and a major leap toward personal superintelligence,” states Meta’s blog post regarding the new technology.
The operational scale of these models is reflected in their pricing structure. Available through the Meta Model API, Wonder is priced at $4.00 per million input tokens and $20.00 per million output tokens, positioning it roughly alongside models like GPT-6 Sol or Opus 5.5 in cost.
It is important to note that Wonder Full will be restricted to Muse Code subscribers and will not be accessible via the standard API. Both models are intended to serve as the foundation for the Muse agent.
Aligning with Meta's goal of creating persistent personal assistants, the company highlighted the new models' ability to retain and synthesize user information, a capability termed “active memory.”
To validate this memory capability, Meta utilized the MemoryBench benchmark, which assesses continual learning frameworks. The testing indicates that Muse Wonder now outperforms Fable, Opus, and Astra on this metric.
“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,” explained research scientist Kavya Kopparapu in a video accompanying the launch.
These models represent the first frontier model release following the restructuring of Meta's AI efforts by Scale AI founder Alexandr Wang last August. Since then, Meta has primarily focused on lightweight releases like Muse Spark and the Muse agent, avoiding the high-profile state-of-the-art releases typical of competitors like OpenAI and Anthropic.
The post specifies that the models were trained at the Prometheus data center in central Ohio, which became operational earlier this year.
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.