Meta Launches Muse Wonder, Its First Frontier AI Model, With Split Access Routes

Meta released Muse Wonder on October 12, 2026, its first frontier large language model, developed by Meta Superintelligence Labs and trained at a gigawatt-scale data center called Prometheus. The model comes in two variants aimed at long-running agentic tasks, with features such as Active Memory for retaining context across sessions and Active Focus for persisting through obstacles. Meta claims state-of-the-art results on agent, memory, and coding benchmarks, though these figures are vendor-reported and not yet independently confirmed. Access differs by variant: the standard model is available through Meta's API with reported per-token pricing, while the stronger Wonder Full version is currently limited to Muse Code subscribers. Developers have reported confusion over these access differences and setup issues since the launch.