Meta’s Superintelligence Lab announced two new large models on Monday: Muse Wonder and Muse Wonder Full. The company describes these as its first frontier-class releases.
Compared with the earlier Muse Spark, which targeted low latency and lightweight deployment, the Wonder models are larger and intended to be more capable. Meta’s published model card reports that Muse Wonder and Muse Wonder Full reach benchmark scores comparable to Fable 5.1 and GPT‑6 Astra, and produce stronger results on complex tasks such as coding.
Meta framed the release as a step in its scaling strategy: “Together, they mark the next step on our scaling ladder and a major leap toward personal superintelligence,” the company wrote in its blog post.
Pricing and availability
Muse Wonder will be available through the Meta Model API at a price of $4.00 per million input tokens and $20.00 per million output tokens, a cost level Meta says is roughly equivalent to GPT‑6 Sol or Opus 5.5. Muse Wonder Full, by contrast, will only be available to Muse Code subscribers and will not be served via the API. Both models will also be used to power the Muse agent.
Focus on long-term memory
Meta emphasized the models’ ability to retain and synthesize user-provided information over time, which it refers to as “active memory.” To demonstrate improved recall, the company highlighted results on the MemoryBench benchmark, which evaluates continual learning and memory construction frameworks. According to Meta’s testing, Muse Wonder scores higher on MemoryBench than Fable, Opus, or Astra.
In an accompanying video, research scientist Kavya Kopparapu said: “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.”
Development context
These are Meta’s first frontier model releases since Alexandr Wang, founder of Scale AI, restructured Meta’s AI efforts in August of last year. After that reorganization, Meta released the lightweight Muse Spark model and the Muse agent, but had not previously published frontier-level state-of-the-art models on the scale typical of some competitors.
Meta also said the models were trained at its Prometheus data center in central Ohio, which came online earlier this year.
Why this matters
The new models signal Meta’s move toward offering more capable, longer-memory personal assistants and greater participation in frontier-model development. Product availability and pricing choices — API access for Wonder versus the subscriber-only Wonder Full — will shape how developers and enterprises adopt the models.
Key facts
- Models: Muse Wonder and Muse Wonder Full (Meta Superintelligence Lab).
- Performance: Meta reports scores comparable to Fable 5.1 and GPT‑6 Astra; higher results on complex tasks and MemoryBench.
- Pricing (Muse Wonder, Meta Model API): $4.00 per million input tokens; $20.00 per million output tokens.
- Availability: Wonder via API; Wonder Full limited to Muse Code subscribers and not available via API.
- Use: Both models will power the Muse agent.
- Training site: Prometheus data center, central Ohio (came online earlier this year).
- Background: First frontier release after Meta’s AI reorganization under Alexandr Wang last August.
