Meta Launches Muse Wonder as First Frontier AI Model

According to a report by TechCrunch, Meta’s Superintelligence Lab released Muse Wonder and Muse Wonder Full on Monday, marking the company's first attempt at a frontier model. Unlike the earlier lightweight Muse Spark, these new larger models delivered higher benchmark scores and improved performance on complex tasks like coding. Meta's testing showed these systems achieved scores comparable to Fable 5.1 and GPT-6 Astra. Trained at the Prometheus data center in central Ohio, Wonder is priced at $4 per million input tokens and $20 per million output tokens via the Meta Model API, while Wonder Full is restricted exclusively to Muse Code subscribers without API access. Both systems will power the Muse agent, utilizing a feature called active memory to retain and synthesize user information over extended periods, scoring higher on the MemoryBench evaluation framework than competing options from Fable, Opus, or Astra.

This release represents a significant shift for Meta under the restructured artificial intelligence efforts led by Scale AI founder Alexandr Wang. By prioritizing active memory, Meta is targeting long-term task execution for personal assistants, aiming to retain critical context over days, weeks, and months.

Meta’s introduction of the Muse Wonder series establishes a new tier in its artificial intelligence roadmap, balancing advanced coding capabilities with extended memory retention for personal assistant applications.