Meta announced the debut of two new artificial intelligence models, Muse Wonder and Muse Wonder Full on Monday. The company stated that it’s the most advanced model yet, according to The New York Times.
Muse Wonder and Muse Wonder Full are on par with competitor Anthropic with advanced models on testing, reasoning, writing and coding according to The New York Times. These results have been provided by Meta and have not been independently verified.
Meta has stated that Muse Wonder is the first model that is competing with “frontier” AI models, which are powerful systems and usually require the most computing power to run on, Times reported.
Users of Meta’s Muse personal agent app will need a monthly subscription of $20 or $100 to access Muse Wonder, according to CNBC. They reported how the release is part of Meta’s effort to build AI revenue outside its online advertising business.
Muse Wonder Full, the more powerful version, will only be available to subscribers of Muse Code, which is Meta’s AI coding tool, according to Times. CNBC reported those plans cost $50 or $200 per month.
Muse is an AI agent, which means it can operate apps and software on a user’s behalf. Its reached the top of Apple’s App Store after launching last month, according to both Times and CNBC.
Alexandr Wang, Meta’s chief AI officer, told CNBC the company trained the model so agents don’t stall on tasks such as booking dinner reservations. He mentioned that the model was built to “try other approaches instead of stalling,” when it reached obstacles.
This new launch also comes as concerns grow over advanced AI agents, some of which have hacked into companies and altered government websites, according to Times. Blog posts from Meta have stated that Muse Wonder has its strongest safeguards against cyber misuse so far, CNBC reported. These protections come only with the paid model.
Some Muse users have complained on social media that the agent accessed their personal messages without permission, according to CNBC. Meta has disputed the claims and Wang said users may have given permission earlier in conversation and forgotten. The company is now focusing on educating users.
The models were delayed after Meta paused a stage of development called pre-training in July, according to Times. Meta CEO Mark Zuckerburg teased the release on an Instagram livestream, noting that a “big watermelon” was on the menu while he barbecued. The models were internally codenamed Watermelon.