Meta has introduced Muse Wonder Full, a new AI model that scores 56 in Artificial Analysis’s independent AI model rankings, according to the original report. That puts it ahead of Claude Fable 5.1 and GPT-6 Astra, which both score 53.
Only Claude Opus 5.5 scores higher at 58, while Sonnet 5.5 reportedly ties with Meta’s new model.
The model is the latest release from Meta’s AI lab, which Mark Zuckerberg reportedly rebuilt from scratch over the past 14 months.
What Makes Muse Wonder Different?
Meta designed Wonder Full for complex, long-running tasks, including coding projects and legal paperwork.
Key features include:
- Long-running tasks: Designed to work through complex projects over extended periods.
- Memory across conversations: Meta says it can remember relevant details about users across chats.
- Persistent problem-solving: It is designed to continue working when it encounters obstacles.
- Paid access: The full version is available through Muse Code, Meta’s coding application, starting at $50 per month.
- Developer access: Developers can access the standard Wonder model at a price comparable to Claude Opus 5.5.
- No local downloads: Neither version can be downloaded and run locally, unlike Meta’s earlier Llama models.
Privacy Concerns and What Comes Next
The announcement also raises questions about data privacy. The original report references a recent TIME investigation into Muse’s handling of files involving users and their contacts.
As AI models become more capable and retain more context across conversations, users will need greater transparency about how their data is stored and used.
Note: Verify the model names, benchmark scores, pricing, and privacy claims against reliable primary sources before publishing, as these details are taken from the supplied newsletter text.
