How Meta launched Muse Wonder on X

Muse Wonder and Muse Wonder Full are Meta models for agents that take on multi-step personal and professional tasks, including insurance claims, lease reviews, finance and health questions, and longer coding projects. Developers can use Muse Wonder today through the Meta Model API and Muse Code, while Muse Wonder Full is available only to Muse Code subscribers. Neither model is being released with downloadable weights. The models will soon power Meta's Muse personal agent, rolling out first to subscribers. The October 12, 2026 release comes from Meta Superintelligence Labs, led by Chief AI Officer Alexandr Wang, and follows a rebuild of Meta's AI work that began fourteen months earlier. It is the first set of models from that lab that Meta is willing to compare directly with the best models from OpenAI and Anthropic. Active Memory carries preferences, habits, and other useful context across conversations and brings those details back when a task needs them. Active Focus is trained to continue when work stalls, try other approaches, recover from errors, and check in before important actions. The models accept text, image, video, and audio across a context window of about 1 million tokens. On the Meta Model API, Muse Wonder costs $4 per million input tokens and $20 per million output tokens, with cached input at $0.20 per million. Both models were trained at Meta's Prometheus data center.