Meta has now launched the most powerful AI models to date: Muse Wonder and Muse Wonder Full. Muse Wonder is available today for developers via the Meta Model API and Muse Code, and will be available later as an app, beginning with subscribers. The app is not available outside of the U.S. and Canada. What the benchmarks reveal and what it means for Europeans.
On Oct. 12, Meta released two new language models. Muse Wonder and the larger Muse Wonder Full are built upon Muse Spark, and are designed for agents or AI doing tasks for you. Meta has examples such as dealing with an insurance claim, comprehending a lease and creating a working application. The company emphasizes two aspects. Active Memory is designed to transfer relevant information between conversations, while Active Focus maintains the model’s momentum towards a goal when a task becomes challenging, attempts alternative strategies and revisits the model before significant points. Both models have a context window of 1 million tokens, according to Decrypt.
Who can use Muse Wonder today.
According to Meta, Muse Wonder is now available to developers and builders via the Meta Model API and Muse Code, its coding agent. Muse Wonder Full is only available to Muse Code subscribers. Meta’s personal AI agent, the Muse app, is yet to get the new model. What can be said of Muse Wonder is that it will be “soon” available in Muse, and will be available first to subscribers. No date is shown for free users, nor are the countries where the API is available listed.
The application is currently only available in North America. Muse made its debut in the U.S. in September and hit Canada in mid-September. Europe is yet to receive a date from Meta. Meta isn’t the sole firm to hold off on bringing their agent to Europe. On Oct. 12, Google unveiled its Gemini Spark agent as a Task button in the main Gemini chat. Spark is not included on the European Economic Area, Switzerland, the United Kingdom and Nigeria pages of the Gemini Apps Help page.
What the benchmarks show
Meta primarily references its own statistics. Muse Wonder Full outperforms Claude Opus 5.5 with 59.1 on Meta’s in-house Personal Superintelligence Index, while GPT-6 Astra trails with 54.5. Meta has not published how the index is calculated. The company also claims its SWE-Marathon coding comparison was not controlled and its model ran at full throttle.
The models are ranked lower by independent testing. Claude Opus 5.5 is the highest-scoring AI tool with a rating of 58 on the Artificial Analysis Intelligence Index, which is a weighted average of 10 evaluations. Muse Wonder Full is tied for 56 points with Claude Sonnet 5.5. The average performance of the standard Muse Wonder developers can purchase is 52, 1 point lower than GPT-6 Astra and Claude Fable 5.1. That puts them in the top scoring position but not on top.
Pricing and privacy
Muse Wonder costs $4 per million input tokens and $20 per million output tokens through the API. That’s the same price as Claude Opus 5.5, OpenAI’s $10 and $50 for GPT-6 Astra, Decrypt reports. There are other free sites that offer people the ability to chat without paying. Recently, ChatGPT has introduced GPT-6 Luna for free.
Agents have access to email, calendars and documents, and Meta says it trained the models to avoid sharing private information without permission. A full safety report will be available later, no date given. In Europe, users have the right to refuse to have their content used for AI training based on their public posts on Facebook and Instagram. Our guide on how to opt out of AI training explains how to do it.
Source(s)
Meta: Introducing Muse Wonder and Muse Wonder Full
Decrypt: Meta’s New Muse Wonder Finally Puts Zuckerberg in the Frontier AI Race
