Meta unveils two new AI models designed to handle complex tasks

Yesterday 22:42

By: Azzat Manal

Meta unveils two new AI models designed to handle complex tasks

Meta has announced two new artificial intelligence models, Muse Wonder and Muse Wonder Full, which the company describes as its most capable and efficient systems to date. The models are designed to move beyond answering questions and generating content, with a focus on carrying out complex tasks for both personal and professional use.

The launch follows an extended restructuring of Meta’s artificial intelligence strategy, which began around 14 months ago and included work on the Muse Spark model. The latest systems are intended to support practical activities such as reviewing insurance claims, analyzing financial information, developing software and building applications.

A central feature of the new models is what Meta calls active memory, a capability designed to retain relevant information across conversations and retrieve it when needed. This could help users manage longer projects without repeatedly providing the same background information, although the usefulness of such a feature will depend on how accurately information is stored and applied.

The models also incorporate an approach described as active focus, allowing them to continue working toward a defined objective, test alternative solutions and correct errors along the way. According to the company, they can seek user confirmation before taking important actions, an important safeguard as AI systems become more involved in tasks with real-world consequences.

Meta is targeting applications in areas including law, finance, healthcare, programming and application development. In these fields, AI systems can potentially reduce repetitive work and help users process large volumes of information. However, professional deployment also requires reliable results, strong data protection and appropriate human oversight, particularly when decisions could affect people’s finances, legal rights or health.

Access to the two models will differ. Muse Wonder is available to developers through Meta’s Model API and the Muse Code service, while Muse Wonder Full is initially restricted to Muse Code subscribers. Meta has also indicated that it plans to bring the first model to experiences on the Muse platform in the near future.

Security and privacy are among the other areas highlighted by the company. Meta says the models were trained to avoid sharing personal information without authorization and to reject requests that could facilitate harmful activities. The company also points to potential cybersecurity applications, including identifying software vulnerabilities and helping improve the protection of digital systems.

These capabilities reflect a broader shift in the AI industry toward systems that can perform multistep tasks rather than simply respond to individual prompts. Such tools may offer productivity gains, but they also introduce challenges involving accuracy, access permissions, accountability and the risk of unintended actions.

Meta says the models were developed in line with its framework for scaling superintelligence. Their practical performance, however, will ultimately depend on how well they handle demanding tasks outside controlled testing environments and how effectively their safeguards operate in everyday use.

The release marks another step in Meta’s efforts to compete in an increasingly crowded AI market, where technology companies are investing in systems capable of reasoning, using tools and supporting more complex workflows. As access expands, developers and users will be watching whether the new models can deliver measurable improvements while maintaining the reliability and security required for sensitive applications.