Meta has launched two new artificial intelligence models as the technology giant intensifies its efforts to compete with leading AI developers, including OpenAI, Anthropic and Google.
The company announced on Monday that its new models, Muse Wonder and Muse Wonder Full, are its most capable AI systems to date, designed to support personal and professional AI agents capable of handling complex tasks with greater independence.
According to Meta, the models can perform long-running tasks, operate computers and assist users in specialised fields such as law, finance and healthcare.
They are also designed to manage complex software development projects, delegate tasks to smaller AI agents and recover from errors without abandoning their original objectives.
The launch represents another step in Meta’s broader strategy to develop AI systems that can do more than respond to prompts, with an emphasis on completing practical tasks and adapting to users’ needs.
New Features Target Personalised AI Assistance
Among the key features introduced with the Wonder models is Active Memory, which enables the AI systems to retain important information about users across multiple conversations and retrieve it when relevant.
The feature is intended to make interactions more personalised by allowing the models to draw on previously shared information rather than treating every conversation as an entirely separate exchange.
Another feature, Active Focus, is designed to help AI agents continue working when they encounter obstacles.
Meta said the capability allows the models to explore alternative approaches, adapt to new information and seek clarification when necessary instead of abandoning a task after encountering difficulties.
The company said the models could support a wide range of everyday and professional activities, including reviewing rental agreements, navigating insurance claims, analysing personal finances and developing software applications.
These capabilities reflect the growing industry focus on AI agents that can plan and execute multistep tasks with limited human intervention.
Availability and Subscription Plans
Meta said Muse Wonder is now available to developers through its Meta Model API and Muse Code programming tool.
The more advanced Muse Wonder Full model is available through paid Muse Code subscriptions, with plans starting at $50 per month for access to the company's most advanced models.
Meta also plans to integrate Muse Wonder into its Muse personal assistant at a later date, initially making the feature available to subscribers.
The release provides developers with access to the models for building applications and experimenting with AI-powered workflows, while the subscription offering targets users seeking more advanced coding and productivity capabilities.
Meta Claims Competitive Benchmark Performance
Meta published benchmark results that it said showed the Wonder models matching or outperforming selected systems developed by OpenAI and Anthropic in areas including coding, memory and professional tasks.
However, the results represent the company's own assessment of its models, and their broader performance will require further evaluation through independent testing.
Performance can vary depending on the task, testing methodology and conditions under which competing systems are assessed. Independent evaluations will therefore be important in determining how the new models compare across different real-world applications.
Safety and Privacy Measures
Meta said it trained the Wonder models to protect private information, refuse requests involving harmful activities and strengthen their defences against cyberattacks.
The company indicated that a more detailed safety report would be released later, providing additional information about the models' safeguards and risk-management measures.
The safety of increasingly autonomous AI agents has become an important issue across the technology industry, particularly as such systems gain the ability to operate computers, handle sensitive information and carry out complex tasks.
Zuckerberg's Vision of Personal Superintelligence
The launch forms part of Meta's wider push towards what Chief Executive Officer Mark Zuckerberg has described as personal superintelligence — AI systems designed to understand individual users' goals, preferences and long-term needs.
The concept centres on developing AI assistants capable of providing more personalised support across work, communication, planning and other aspects of daily life.
Meta's latest models are intended to advance that vision by combining memory, problem-solving and task execution in systems designed for both developers and everyday users.
Notably, the company did not announce plans to release the Wonder models' downloadable weights as open-source software, unlike some of its earlier AI models.
The decision leaves questions about the extent to which outside developers and researchers will be able to inspect, modify or independently deploy the new systems.
As competition in the AI sector intensifies, the performance, accessibility, safety and practical usefulness of Meta's latest models will help determine their position in a market increasingly focused on AI agents capable of completing complex tasks rather than simply generating text and answering questions.
