Zuckerberg says Meta has reached AI ‘frontier’

Meta unveiled two new frontier AI models, Muse Wonder and Muse Wonder Full, on Monday, marking the company’s first release of large‑scale, high‑performance models. The models, accessed via the Meta Model API, are positioned as more capable successors to the lightweight Muse Spark, delivering higher benchmark scores and stronger performance on complex tasks such as coding. Meta’s internal testing, detailed in the models’ cards, shows Wonder and Wonder Full achieving results comparable to leading competitors like Fable 5.1 and GPT‑6 Astra, with pricing set at $4 per million input tokens and $20 per million output tokens—roughly on par with GPT‑6 Sol or Opus 5.5.

A central focus of the Wonder series is “active memory,” Meta’s term for the ability to retain and synthesize user‑provided information over extended periods. The company highlighted superior scores on its MemoryBench benchmark, which evaluates continual learning and memory construction, claiming Muse Wonder outperforms rivals such as Fable, Opus, and Astra in this metric. Research scientist Kavya Kopparapu emphasized the effort put into enabling the model to remember important user data across days, weeks, and months, a capability Meta sees as essential for personal assistants that operate over long time horizons.

The launch follows a restructuring of Meta’s AI division led by Scale AI founder Alexandr Wang in August 2023, after which the firm released the lightweight Muse Spark and a Muse agent but had avoided flagship model releases like those from OpenAI and Anthropic. The new models were trained at Meta’s Prometheus data center in central Ohio, which became operational earlier this year. By introducing frontier models with advanced memory features, Meta signals a strategic push toward “personal superintelligence,” aiming to compete directly with leading industry models while expanding its AI service offerings through the Model API.

⚡ Effects Interpreter