Meta says its new Muse Wonder AI has caught OpenAI and Anthropic at the frontier

Fourteen months after tearing up its AI work and starting again, Meta says it has caught the leaders. The company unveiled Muse Wonder and Muse Wonder Full on Monday, calling them its most capable models yet and claiming state-of-the-art results on personal and professional agent tasks, ahead of OpenAI’s and Anthropic’s best on several of its own charts.

Mark Zuckerberg told the Financial Times that Meta has caught up with rivals at the frontier of AI, and rejected calls from other AI bosses for an industry slowdown. On X, he put it more simply:

It’s been just over a year since we started Meta Superintelligence Labs. Muse Spark was our first family of models, and now Muse Wonder is a much larger and more advanced class of models that brings us to the frontier.

We’re releasing two models today. Muse Wonder for fast frontier performance and Muse Wonder Full for the full version of the model.

— Mark Zuckerberg (@finkd) October 12, 2026

What Muse Wonder does

Meta built the two models for agents: AI that carries out tasks for you rather than only answering questions. Its research blog says they were trained at Prometheus, its gigawatt-scale data centre, and lean on two new abilities:

• Active Memory: carrying what matters about you across conversations and pulling back the details relevant to the job in hand.
• Active Focus: keeping going when a task gets hard, trying other approaches, recovering from errors, asking for clarification and checking in before doing anything important.

Meta also pitches the models at expert work in law, finance and health (reading a lease, an insurance claim or someone’s finances) and at coding, from small apps to long software projects. Its demos show Muse Wonder planning a run in a browser, ordering custom merchandise from a person’s Instagram photos and building a 3D kitchen game.

Alexandr Wang, Meta’s AI chief, told CNBC the model, codenamed Watermelon, was trained so agents don’t stall on everyday jobs such as booking a dinner reservation:

When you do real-world tasks, you’re going to get roadblocks and you’re going to get things that come up or various setbacks. We specifically tried to train the model to be able to try other approaches instead of stalling and then checking back in with the user at any time.

Alexandr Wang, Meta’s AI chief, to CNBC

How it compares, by Meta’s numbers

The benchmark charts are Meta’s own and have not been independently checked. On MemoryBench, an internal test of how well a model remembers a user’s context over a long history, Meta scores Muse Wonder at 85.6% against 81.5% for Claude Opus 5.5 and 81.4% for GPT-6 Astra. On AutomationBench-AA, which tests end-to-end business workflows, Muse Wonder Full leads with 45.4%, while the smaller Muse Wonder (39.1%) trails both GPT-6 Astra (41.6%) and Claude Opus 5.5 (41.2%).

Meta says the models are its most aligned to date, were trained not to share private information without permission, and carry its strongest safeguards yet against chemical, biological and cyber misuse. A full safety report is promised “soon”. That matters for an agent that reads messages, calendars and documents: Muse users have already complained of the agent acting on their behalf without telling them.

Who can use it, and what it costs

Muse Wonder is available now through the Meta Model API and Muse Code, Meta’s coding agent. Muse Wonder Full is only for Muse Code subscribers, whose plans cost $50 or $200 a month, according to CNBC. On OpenRouter, Muse Wonder is listed at $4 per million input tokens and $20 per million output tokens, with a context window of about one million tokens. Wang told CNBC the API price will still be “meaningfully below” leading models from OpenAI and Anthropic.

For everyone else, Muse Wonder will “soon” power Muse, Meta’s personal agent app, starting with paying subscribers on its $20 or $100 a month plans. Zuckerberg says Meta will “work to bring this level of intelligence to everyone over time”, and Meta plans to use Muse Wonder to build smaller, specialised models for free users. Meta’s announcements don’t list countries, so availability should follow the Muse app and Meta Model API, which don’t yet cover every market. Developers can already try it in Meta’s $1 million global hackathon.

Why it matters

Meta’s earlier Muse Spark models were cheap and capable, and it even released Muse Glimmer to run locally, but none topped the leaders. If Muse Wonder’s numbers hold up in outside tests, OpenAI, Anthropic and Google have a fourth rival at the frontier, with billions of users to sell it to.

Sources: Meta announcement; Meta research blog: Muse Wonder; Mark Zuckerberg on X; CNBC interview with Alexandr Wang; Financial Times interview with Mark Zuckerberg; OpenRouter: Muse Wonder.