MENLO PARK, CALIFORNIA / RankWire.AI / – Meta has rolled out Muse Wonder and Muse Wonder Full, its most advanced AI models to date. Designed specifically for programming, persistent AI agents, and complex personal or professional tasks, these systems aim to enhance user capabilities. Meta states that these models can assist with managing insurance claims, interpreting leases, analyzing financial data, and developing functional applications. On October 12, Muse Wonder became accessible through the Meta Model API and Muse Code, while Muse Wonder Full was launched exclusively for Muse Code subscribers.
This launch marks a significant step in Meta’s expansion into AI systems capable of executing multi-step processes, moving beyond simple prompt responses. Muse Wonder can plan extensive tasks, assign subtasks to sub-agents, utilize software tools, and recover from errors during execution. Additionally, Meta has optimized these models for advanced computer use across browsers, desktop software, and mobile apps. Both models support a context window of nearly one million tokens, enabling them to process large data volumes within a single interaction.
In support of longer, more tailored interactions, Meta introduced two features: Active Memory and Active Focus. Active Memory retains pertinent information across conversations and retrieves details when relevant to a new task. Active Focus helps the models maintain progress when a task becomes challenging, try alternative strategies, and recover after errors. Meta also highlighted that these models can request clarification from users and verify actions before execution.
Focus on coding and persistent AI agent functionalities
Pricing for Muse Wonder is set at $4 per million input tokens and $20 per million output tokens via the Meta Model API. The cost for cached input tokens is $0.20 per million, as per the company’s developer resources. Instead of separate API pricing, Muse Wonder Full is available only through paid Muse Code subscriptions. Developers with higher workloads can opt for Power Usage at $50 a month and Maximum Usage at $200 per month.
Meta reports that Muse Wonder Full outperforms many benchmarks related to agentic and coding capabilities, with results published alongside the release. Tests like DeepSWE and SWE-Marathon, which assess software engineering and long-term coding efficiency, were used for evaluation. Independent reviewer Artificial Analysis assigned Muse Wonder an Intelligence Index score of 52 and Muse Wonder Full a score of 56 at maximum reasoning effort, placing these models above Meta’s earlier Muse Spark generation on the evaluator’s index.
Enhanced safety measures accompany expanded model features
Meta trained the Muse Wonder series to avoid sharing private data without consent and to refuse requests involving harmful activities. The models also retain cybersecurity functionalities, aiding developers in detecting vulnerabilities in their code and reinforcing security defenses. The company states that the models are evaluated under the Meta Superintelligence Scaling Framework and plans to publish a comprehensive safety report that covers additional safety assessments following the initial launch.
Approximately 14 months prior to the Muse Wonder debut, Meta embarked on a major overhaul of its AI initiatives. The new models are built upon the foundations laid by Muse Spark, with training conducted at the Prometheus data center. The company intends to expand Muse Wonder’s integration into more features within its Muse platform, initially targeting subscribers. The October rollout provides immediate access to Muse Wonder for developers, while the more comprehensive Muse Wonder Full remains exclusive to Muse Code subscribers.