Meta releases Muse Wonder, its first frontier AI model built to rival GPT and Claude

Meta releases Muse Wonder, its first frontier AI model built to rival GPT and Claude

Meta has released its first frontier AI model, Muse Wonder, alongside a larger variant called Muse Wonder Full, in what the company describes as a major leap toward its goal of building personal superintelligence.

The Wonder models are the most powerful Meta has ever released, reaching benchmark scores it says are comparable to Anthropic's Fable 5.1 and OpenAI's GPT-6 Astra. Where Meta's earlier Muse Spark was built for low-latency lightweight tasks, the Wonder models are designed for complex, multi-step work including coding, reasoning, and extended memory tasks.

A key differentiator Meta is emphasising is what it calls "active memory": the ability to retain and synthesise personal information provided by users across days, weeks, and months. According to Meta's internal testing, Muse Wonder scores higher on the MemoryBench benchmark than any model from Anthropic, OpenAI, or Google.

Accessed through the Meta Model API, Wonder will cost $4 per million input tokens and $20 per million output tokens, making it broadly equivalent in price to GPT-6 Sol or Opus 5.5. Wonder Full will be available only to Muse Code subscribers and will not be served via the API. Both models will power Meta's Muse agent.

The models were trained at Meta's Prometheus data centre in central Ohio. They represent the first major frontier model release since Scale AI founder Alexandr Wang restructured Meta's AI strategy in 2025.