Meta published an official announcement introducing Muse Wonder and Muse Wonder Full, describing them as its most capable models to date for personal and professional agentic tasks. The company says Muse Wonder is available immediately through the Meta Model API and Muse Code, while Muse Wonder Full is limited to Muse Code subscribers.
A companion Meta AI Research post frames the release as the next step after Muse Spark on Meta's scaling ladder. Both posts stress Active Memory, which carries important user context across conversations, and Active Focus, which keeps working through hard tasks, tries alternatives, and checks in before consequential actions.
Meta says the models target expert domains such as law, finance, and health, and claim frontier coding performance for building apps and games. Training happened at Meta's gigawatt-scale Prometheus data center. The company plans to use Muse Wonder to power Muse experiences first for subscribers, and later to derive specialized models for broader Muse users.
Independent coverage from TechCrunch, The Verge, and Reuters confirmed the same-day launch and Meta's claim that Wonder is its first clear frontier-model bid since Alexandr Wang restructured Superintelligence Labs. TechCrunch cited API pricing of about $4 per million input tokens and $20 per million output tokens for Wonder, and noted Wonder Full is not served on the public API.
Meta's posts highlight safety work under its Superintelligence Scaling Framework, including stronger resistance to private-data leakage and misuse while preserving defensive cyber use for developers. A fuller safety evaluation report is promised soon. Benchmark comparisons to other frontier models appear in Meta's research materials and should be treated as vendor-reported until third parties reproduce them.
Availability is gated: API and Muse Code access for Wonder, subscriber-only access for Wonder Full, and a staged roll-out into the Muse agent. That packaging signals Meta is tying frontier capability to paid developer and subscriber surfaces rather than a free default chat model.
For builders, the practical change is a new Meta Model API endpoint family plus Muse Code access today. For the broader Muse app audience, the models matter when subscriber roll-out begins and when specialized derivatives ship.
Taken together, the release shows Meta competing on long-horizon agent reliability and paid distribution rather than on a free general chat default. Developers can start evaluating Wonder through documented API and Muse Code paths while waiting for independent benchmarks and the promised safety report.
Enterprise buyers should still separate the product announcement from unverified leaderboard claims. Until outside evals land, the durable facts are availability channels, subscriber gating for Wonder Full, and Meta's agent-first positioning around memory and sustained focus.