MENLO PARK, CALIFORNIA / RankWire.AI / – Meta has unveiled its new family of sophisticated AI models, named Muse Wonder and Muse Wonder Full. These systems are designed for tasks involving software development, AI agents, and complex daily activities. They can assist with processes such as leases, insurance claims, financial data, and application creation. Meta announced the release of Muse Wonder via its Model API and Muse Code on October 12, with Muse Wonder Full becoming accessible to users holding eligible Muse Code subscriptions.
The models are optimized for multi-step tasks rather than simple, single-response queries. Muse Wonder is capable of planning workflows, utilizing digital tools, and breaking down larger projects into smaller, specialized sub-agents. It is compatible across browsers, desktop programs, and mobile platforms. The company designed the models to recover seamlessly if a selected method encounters failure. Supporting context windows of approximately one million tokens, these models can process extensive documents, codebases, and large-scale materials within a single session.
Meta also introduced features called Active Memory and Active Focus alongside the Wonder models. Active Memory stores pertinent details from prior conversations and retrieves them when necessary. Active Focus enables an AI agent to persist through challenging problems and adjust its approach following mistakes. The models can ask for clarification if instructions lack crucial information and can seek confirmation before executing sensitive or critical actions. These enhancements aim to facilitate longer work periods without the need for users to restart tasks repeatedly.
Enhanced agent and coding functionalities introduced with new models
The cost of using Muse Wonder is $4 per million input tokens via the Meta Model API. Output generation costs $20 per million tokens, while cached input data is priced at $0.20 per million tokens. Instead of separate API fees, Meta offers Muse Wonder Full through paid Muse Code plans. The company also provides a Power Usage plan at $50 per month and a Maximum Usage plan at $200 per month, targeting developers with higher demands for coding tasks.
Meta has shared benchmark results assessing software engineering, coding agents, and long-duration technical tasks. These include tests like DeepSWE and SWE-Marathon. Additionally, Artificial Analysis evaluated both models at maximum reasoning effort, assigning Muse Wonder an Intelligence Index score of 52 and Muse Wonder Full a score of 56. These outcomes surpass Meta’s previous Muse Spark models on the evaluator’s scale, reflecting a noticeable improvement in reasoning capabilities during testing.
Safety and privacy measures accompany the new models’ launch
Meta emphasizes that the Wonder models incorporate safeguards to protect privacy, prevent harmful requests, and ensure cybersecurity. The models are trained to avoid sharing private data without user approval and to reject requests involving malicious activities, while still enabling legitimate security functions. Developers can utilize the models to identify vulnerabilities in their own software and strengthen defenses. Meta is evaluating its most advanced AI systems under its Superintelligence Scaling Framework and plans to publish additional information about safety testing for these new models.
The debut follows a restructuring of Meta’s AI initiatives that started approximately 14 months earlier. The company states that Muse Wonder builds upon the technical foundations established through Muse Spark. Development of the new generation utilized computing resources available at its Prometheus data center. The company also intends to incorporate Muse Wonder technology into additional features within the Muse platform, beginning with subscribers. Currently, developers can access Muse Wonder, while Muse Wonder Full remains available through qualifying Muse Code subscription plans.