Meta launches Muse Wonder and Wonder Full as its first frontier AI models

Meta on 12 October 2026 released Muse Wonder and Muse Wonder Full, describing them as its first frontier-class models for personal and professional AI agents, according to Meta’s research blog and TechCrunch.

Availability and pricing

Wonder is available through the Meta Model API and Muse Code. TechCrunch reported API pricing of $4 per million input tokens and $20 per million output tokens. Wonder Full is for Muse Code subscribers only and is not served through the API. Both models will power the Muse agent; Meta says Wonder will later roll into broader Muse experiences for subscribers first.

Capabilities Meta highlights

Meta emphasizes Active Memory (carrying user context across conversations) and Active Focus (sustained work on hard goals with clarification and check-ins). It claims strong results on agent, memory and coding benchmarks — including MemoryBench scores above Fable, Opus and Astra in Meta’s own tests — and comparable overall scores to Fable 5.1 and GPT-6 Astra on its model card. The models were trained at Meta’s Prometheus data center in Ohio.

Safety and caveats

Meta says Wonder has its strongest safeguards yet against CBRN and cyber misuse while allowing defensive cyber use under its Superintelligence Scaling Framework. Benchmark and safety claims are Meta’s own; independent rankings may differ, and Wonder Full’s closed distribution limits third-party API evaluation.