MENLO PARK, CALIFORNIA / RankWire.AI / – Meta has launched its most sophisticated artificial intelligence models to date, named Muse Wonder and Muse Wonder Full. These new systems are engineered for tasks such as coding, managing long-duration AI agents, and handling complex personal or professional activities. Meta stated that these models can assist users in processing insurance claims, understanding lease agreements, analyzing financial data, and developing operational applications. Muse Wonder became accessible on October 12 via the Meta Model API and Muse Code, with Muse Wonder Full debuting for Muse Code subscribers.
This development extends Meta’s efforts into AI systems capable of executing multi-step tasks, moving beyond simple prompt-based responses. Muse Wonder is designed to plan extended sequences, assign work to sub-agents, utilize software tools, and recover from errors during execution. The models are also built for advanced computer use across browsers, desktop software, and mobile apps. Both models feature a context window of roughly one million tokens, enabling them to handle large data volumes within a single session.
Meta introduced two features—Active Memory and Active Focus—to facilitate longer, more personalized interactions. Active Memory maintains relevant information across conversations and retrieves details when they are pertinent to new tasks. Active Focus supports the models in continuing work when faced with difficult tasks, testing alternative methods, and recovering from mistakes. Additionally, Meta states the models can ask users for clarification and verify details before executing critical actions.
Models focus on coding and autonomous AI agents
Through the Meta Model API, Muse Wonder is priced at $4 per million input tokens and $20 per million output tokens. Cached input costs $0.20 per million tokens, according to the company’s developer resources. Muse Wonder Full is available solely via paid Muse Code subscription plans, not separate API charges. For developers with heavier coding demands, Meta offers Power Usage at $50 a month and Maximum Usage at $200 a month.
Meta claims Muse Wonder Full outperforms many benchmarks related to agentic capabilities and coding. The company highlights tests such as DeepSWE and SWE-Marathon, which assess software engineering and long-running coding efficiency. An independent evaluator, Artificial Analysis, assigned Muse Wonder an Intelligence Index score of 52 and Muse Wonder Full a score of 56 at maximum reasoning effort. These results place the models above Meta’s earlier Muse Spark generation on the evaluator’s index.
Enhanced safety features accompany expanded model functionalities
Meta has trained the Muse Wonder series to avoid sharing private information without consent and to decline requests involving harmful activities. The models also retain cybersecurity functions designed to help developers identify vulnerabilities in their code and bolster security defenses. Meta states that its most capable models are evaluated under the Meta Superintelligence Scaling Framework. A more comprehensive safety report, covering additional safety assessments, will be published following the initial launch.
The AI division at Meta underwent a significant overhaul approximately 14 months prior to the Muse Wonder release. The company notes that these models build upon the foundation set by Muse Spark, with training conducted at its Prometheus data center. Meta plans to integrate Muse Wonder into various experiences within its Muse platform, initially targeting subscribers. The October launch provides developers immediate access to Muse Wonder, reserving the larger Muse Wonder Full model for Muse Code subscribers.