MENLO PARK, CALIFORNIA / RankWire.AI / – Meta has unveiled the Muse Wonder and Muse Wonder Full, a new series of sophisticated artificial intelligence models designed for software development, AI agents, and complex daily tasks. These systems are capable of assisting with areas such as leases, insurance claims, financial data, and app creation. On October 12, Meta launched Muse Wonder via its Model API and Muse Code, making Muse Wonder Full accessible to users with eligible Muse Code subscriptions.
Focusing on multi-step tasks rather than single responses, the models can plan workflows, utilize digital tools, and distribute larger jobs among specialized sub-agents. They are capable of functioning across browsers, desktop software, and mobile platforms. Meta engineered the system to recover when a selected approach fails, supporting context windows of approximately one million tokens, enabling processing of extensive documents, codebases, and other large materials within a single session.
Meta also introduced Active Memory and Active Focus alongside the Wonder models. Active Memory retains useful details from previous conversations and retrieves them as needed, while Active Focus helps AI agents persist through difficult problems and adapt their approach after errors. The models can ask for clarification when instructions lack crucial details and seek confirmation before executing sensitive or significant actions. These features aim to facilitate longer work periods without requiring users to restart tasks repeatedly.
Enhanced capabilities for agents and coding
Muse Wonder is priced at $4 per million input tokens via the Meta Model API, with output costing $20 per million tokens and cached input at $0.20 per million tokens. Instead of separate API pricing, Meta offers Muse Wonder Full through paid Muse Code plans, including a Power Usage plan at $50 per month and a Maximum Usage plan at $200 per month, tailored for developers with heavier coding demands.
Meta shared benchmark results covering software engineering, coding agents, and long-term technical tasks, highlighting tests such as DeepSWE and SWE-Marathon. The evaluations, including maximum reasoning efforts, rated Muse Wonder with an Intelligence Index score of 52 and Muse Wonder Full with a score of 56, placing them above Meta’s earlier Muse Spark models and demonstrating a clear improvement in reasoning performance across tests.
Safety and privacy considerations included in launch
Meta states the Wonder models incorporate protections for privacy, harmful requests, and cybersecurity applications, trained to avoid sharing private information without user consent and to reject harmful activity requests while maintaining legitimate security functions. Developers can still leverage these models to identify vulnerabilities in their software and enhance defenses. Meta assesses its most advanced AI systems under its Superintelligence Scaling Framework and plans to publish additional details regarding safety testing for the new models.
This launch follows a major restructuring of Meta’s AI efforts that began around 14 months earlier. The company states Muse Wonder builds upon technical foundations established through Muse Spark, with the new models trained using computing resources from its Prometheus data center. Additionally, Meta intends to incorporate Muse Wonder technology into more features within the Muse ecosystem, starting with subscribers. While developers can already access Muse Wonder, Muse Wonder Full remains available through qualifying Muse Code subscription plans.