MENLO PARK, CALIFORNIA / RankWire.AI / – Meta has launched Muse Wonder and Muse Wonder Full, a new family of advanced artificial intelligence models. The company built the systems for software development, AI agents and complex everyday tasks. They can assist with areas such as leases, insurance claims, financial information and application development. Meta released Muse Wonder through its Model API and Muse Code on October 12. Muse Wonder Full became available to users with eligible Muse Code subscriptions.
The models focus on tasks that require several steps instead of a single answer. Muse Wonder can plan work, use digital tools and divide larger jobs among specialized sub-agents. It can also operate across browsers, desktop software and mobile applications. Meta designed the system to recover when a chosen approach fails. The models support context windows of about one million tokens, which allows them to process large documents, codebases and other extensive material during one session.
Meta also introduced Active Memory and Active Focus with the Wonder models. Active Memory keeps useful details from earlier conversations and retrieves them when needed. Active Focus helps an AI agent continue working through difficult problems and change its approach after errors. The models can request clarification when instructions lack important details. They can also seek confirmation before carrying out sensitive or important actions. These tools aim to support longer work sessions without requiring users to restart tasks repeatedly.
New models expand Meta’s agent and coding capabilities
Muse Wonder costs $4 per million input tokens through the Meta Model API. Output costs $20 per million tokens, while cached input costs $0.20 per million tokens. Meta distributes Muse Wonder Full through paid Muse Code plans instead of separate API pricing. The company lists a Power Usage plan at $50 per month. Its Maximum Usage plan costs $200 per month and targets developers who need greater access for heavier coding workloads.
Meta published benchmark results covering software engineering, coding agents and long-running technical tasks. The company highlighted results from tests including DeepSWE and SWE-Marathon. Artificial Analysis also evaluated both models at maximum reasoning effort. It gave Muse Wonder an Intelligence Index score of 52 and Muse Wonder Full a score of 56. Those results place the new systems above Meta’s earlier Muse Spark models on the evaluator’s scale and show a measurable increase in tested reasoning performance.
Privacy and safety checks accompany the release
Meta says the Wonder models include safeguards for privacy, harmful requests and cybersecurity use. The company trained them to avoid disclosing private information without user permission. They also reject requests that involve harmful activity while preserving legitimate security functions. Developers can still use the models to identify weaknesses in their own software and improve defensive systems. Meta evaluates its most advanced AI systems under its Superintelligence Scaling Framework and plans to publish more information about safety testing for the new models.
The launch follows a broader restructuring of Meta’s artificial intelligence work that began about 14 months earlier. The company says Muse Wonder builds on technical foundations developed through Muse Spark. Meta trained the new generation using computing resources at its Prometheus data center. The company also plans to bring Muse Wonder technology into additional features within the Muse product, beginning with subscribers. Developers can already access Muse Wonder, while Muse Wonder Full remains available through qualifying Muse Code subscription plans.