Meta’s Superintelligence Lab has released two new AI models, Muse Wonder and Muse Wonder Full, marking the company’s initial entry into frontier models. These models are larger and more capable than Meta’s previous lightweight Muse Spark, delivering higher benchmark scores and improved performance on complex tasks such as coding.
According to Meta’s internal testing, Muse Wonder and Wonder Full achieve benchmark results comparable to leading models like Fable 5.1 and GPT-6 Astra. The company describes these models as a major advancement toward what it calls "personal superintelligence."
Access to Muse Wonder is provided through the Meta Model API, with pricing set at $4 per million input tokens and $20 per million output tokens, aligning its cost with models such as GPT-6 Sol and Opus 5.5. Muse Wonder Full, however, is exclusively available to Muse Code subscribers and is not accessible via the API. Both models will also support the Muse agent, Meta’s personal assistant framework.
A key feature emphasized by Meta is the models’ enhanced ability to retain and synthesize user information over extended periods, a capability the company refers to as "active memory." This is demonstrated by Muse Wonder’s superior performance on the MemoryBench benchmark, which tests continual learning and memory construction, outperforming competitors like Fable, Opus, and Astra.
Research scientist Kavya Kopparapu highlighted the effort invested in ensuring the models can remember important user information over days, weeks, or even months. This focus aligns with Meta’s broader goal of developing AI assistants capable of long-term task management.
These releases follow a strategic restructuring of Meta’s AI initiatives led by Scale AI founder Alexandr Wang in August 2023. Since then, Meta has launched the Muse Spark model and Muse agent but had not introduced frontier models until now.
The training of Muse Wonder and Muse Wonder Full took place at Meta’s Prometheus data center in central Ohio, which became operational earlier this year, underscoring Meta’s investment in dedicated AI infrastructure.