Meta launches Muse Wonder AI models in latest push to rival OpenAI and Anthropic

Meta has introduced two new artificial intelligence models, Muse Wonder and Muse Wonder Full, in its latest effort to compete with leading AI developers including OpenAI and Anthropic. The company says the models are designed to handle demanding personal and professional tasks, with improvements aimed at helping AI agents remember relevant information and stay focused on users' goals.

The launch is part of Meta's wider push to develop AI systems that can do more than answer individual questions. Through its Muse platform, the company is working towards assistants that can help users complete multi-step activities, manage projects and carry out practical tasks with less repeated instruction.

What are Muse Wonder and Muse Wonder Full?

Meta describes Muse Wonder as a model built to support personal and professional agent-based work. Its features include what the company calls Active Memory and Active Focus, intended to help the system retain relevant context and concentrate on the task a user wants completed.

Muse Wonder is available through Meta's Model API and Muse Code, while Muse Wonder Full is available to Muse Code subscribers. Meta says Wonder will also power features in its Muse personal AI agent, with the initial rollout planned for subscribers before the company develops specialised models for wider use.

The distinction between the models and the assistant is important. Muse is the product people interact with, while Wonder and Wonder Full are the underlying models designed to provide more advanced capabilities. Meta is positioning the new systems as part of a broader move towards AI that can work through tasks, rather than simply generate a response to a prompt.

Meta claims its models can compete with AI rivals

Meta says the new models deliver strong performance on personal and professional agent tasks, including coding and work requiring memory and sustained attention. The company has presented benchmark results as evidence that its systems can compete with leading offerings from OpenAI and Anthropic.

Those results should be understood as the company's own claims rather than proof that its models are better in every situation. AI benchmarks measure particular tasks under particular conditions, and a model's performance can vary depending on the type of work, the tools available and how the system is evaluated.

For Meta, the launch is an important test of whether its investment in AI research can translate into practical advantages for users. More capable models could make Muse useful for a wider range of activities, but their long-term value will depend on how reliably they perform outside controlled tests.

What the launch means for AI assistants

The development reflects a wider shift towards AI agents that can help people complete multi-step tasks. Instead of asking a chatbot a question and receiving a single answer, users increasingly want systems that can work towards a goal, remember relevant information and coordinate different actions.

That approach also raises questions about privacy, security and user control, particularly when an AI assistant needs access to personal information or connected services. Meta has said its wider Muse product is designed with security and privacy in mind, but users will still need clear information about permissions and how their data is handled as new capabilities become available.

Muse Wonder and Muse Wonder Full represent another step in Meta's attempt to compete at the leading edge of AI development. The next test will be whether the models deliver reliable results in everyday use and whether developers and subscribers find their capabilities compelling enough to adopt them over rival systems.