“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,” Wang said. “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.”
Meta says Muse Wonder can support tasks across areas including law, finance, health and coding. Examples cited by the company include working through insurance claims, interpreting leases and building apps.
Two features are central to that approach. Active Memory is designed to retain useful information across conversations, while Active Focus is meant to help the model continue pursuing a goal, recover from errors and ask for clarification when necessary.
Muse Wonder will also sit behind paid access in Meta’s consumer AI product. Users of the Muse personal-agent app will need either a $20 or $100 monthly subscription to use the model. Muse itself remains available for free, with subscriptions providing additional usage and access.
Developers can access Muse Wonder through the Meta Model API. Wang said API access will cost more than standard subscriptions because the model is substantially larger and operates at what he described as a frontier level. He also said Meta intends to price it “meaningfully below” leading models from OpenAI and Anthropic.
Muse Wonder Full is more restricted. The model is available through Muse Code, Meta’s software-engineering agent, for users on $50 or $200 monthly subscription plans.
Meta is also emphasizing safety controls around the launch. The company says the Muse Wonder family was trained not to disclose private information without permission, to reject requests involving harmful activity and to support defensive cybersecurity work.
In a blog post, Meta said Muse Wonder “offers our strongest safeguards to date” against “cyber misuse.” Those protections are currently attached to the paid model.
The launch follows the release of Meta’s Muse personal-agent app, which quickly rose in Apple’s App Store. The company has also been monitoring the service for issues as more users test agentic features.
Some users have claimed Muse accessed private messages without authorization, which Meta has disputed. Wang said the company is working to better explain how prior permissions can affect later agent behavior.
“I think there are certainly instances users have authorized various communications or whatnot earlier in a conversation, and then may have forgotten about that,” Wang said. He added that Meta is educating users “to ensure that consumers can trust the product going forward.”
Meta is positioning Muse Wonder as the next step in making its agents more capable of carrying out real-world tasks without stopping whenever something goes wrong. For now, the company is also making its most advanced models and strongest reported safeguards part of its paid AI offerings.
This analysis is based on reporting from CNBC & Meta.
Image courtesy of Meta.
This article was generated with AI assistance and reviewed for accuracy and quality.
