Meta has introduced two AI models – Muse Wonder and Muse Wonder Full. The company describes them as the most powerful to date and a step towards "personal superintelligence." The models are designed for complex tasks: legal work, financial analysis, software development. They handle multi-step projects – from analyzing a lease agreement to creating applications.
"Muse Wonder is a major step towards giving everyone personal superintelligence," said Meta CEO Mark Zuckerberg on X.
In the same post, he provided a comparison: Muse Wonder Full scored 13.1 points in the Harvey LAB-AA legal benchmark against 10.0 for Muse Wonder and 8.6 for GPT-6 Astra. On DeepSWE 1.1 (programming) – 79.0 versus 78.8. On AutomationBench-AA (business processes) – 45.4 versus 39.1.
Meta highlights two features: Active Memory retains information between conversations, Active Focus helps the agent continue working on tasks, try alternative approaches, and recover from errors. The models can request clarifications and coordinate important actions with the user.
Muse Wonder is available to developers through the Meta Model API and Muse Code, Wonder Full – only to Muse Code subscribers. According to the company, the models are trained not to share private information without permission, to refuse harmful requests, and to assist in cybersecurity. Meta promised to release a report with additional security assessments.
The Artificial Analysis platform assigned Wonder Full 56 points on the Intelligence Index, Wonder – 52 at the maximum level of reasoning. This is higher than Claude Fable 5.1 and GPT-6 Astra (both 53), but lower than Claude Opus 5.5 (58). Bindu Reddy from Abacus AI disputed the results, writing on X that Muse Wonder is "nowhere near the level of Fable or Astra."
This material is prepared solely for informational purposes and does not constitute financial advice or a recommendation.
Related articles
