Meta Challenges OpenAI, Anslogic...The highest performance AI 'Muse Wonder' is revealed

Coding, Finance, and Law
Memory of user information for a long time...Performing complex tasks
14 billion investments led by Wang Yi's development
사진 확대 Facebook parent Meta has unveiled a new artificial intelligence (AI) model that will compete with OpenAI's ChatGPT and Anthropic's Claude. It is the highest performance model that Meta CEO Mark Zuckerberg has released since he completely reorganized the AI organization by investing billions of dollars. Meta has entered the competition for AI technology in earnest, claiming that it has secured performance on par with its competitors in coding, finance, and law.
According to Bloomberg on the 12th (local time), Meta unveiled a new AI model called "Muse Wonder." The internal development name is "Watermelon," which is the most performing model developed by Meta so far. Meta explained that the model has significantly improved performance over existing products and has the ability to match the top-level models of OpenAI and Anslogic.
In particular, Meta argues that it has secured competitiveness in the field of writing program codes or analyzing financial and legal issues. However, AI models have large performance differences depending on their work, and test results used for performance evaluation may not accurately reflect the actual use environment. It remains to be seen whether Meta's claims will be confirmed in an external evaluation.
"This model is a true state-of-the-art AI model," Meta Chief AI Officer Alexander Wang said in an interview with Bloomberg. "We can compete with the best models at the moment, and I think it has the highest level of performance, especially in the field of personal and business AI agents." AI agents are technologies that not only seek information under the direction of users, but also perform actual tasks through several stages.
This model is drawing attention in that it is a test to gauge the performance of Zuckerberg's large-scale AI investment and reorganization. Meta started to secure talent and reorganize its research organization after AI models released last year fell short of expectations in the market. Zuckerberg promised to invest hundreds of billions of dollars in building infrastructure such as data centers needed to develop AI models.
What differentiates Meta from its competitors is its personal AI agent function. While competitors such as OpenAI and Anslogic focused on specialized tasks including coding, Meta put weight on developing AI that supports even the daily life of general consumers. Last month, Meta unveiled "Muse," a personal AI agent that handles everyday tasks and errands instead. The recently announced Muse Wonder will soon be applied to Muse's paid subscription service.
Wang explained, "If other companies have focused on coding agents, we have developed them in a differentiated direction so that personal agents can perform well." To this end, Meta has developed an "active memory" that AI remembers information about users for a long time and an "active focus" function that helps them focus on complex tasks.
If there have been cases where existing AI has to explain the user's situation again every time they talk, Muse Wonder can remember and utilize relevant information over several days, weeks, and months. It also has the ability to try other approaches if problems arise during work or if a solution is not found.
"AI can maintain information over days, weeks and months, so users don't have to explain the same thing over and over again," Wang said. This allows us to handle not only simple questions or commands, but also complex tasks where the requirements are not clear.
Following the Muse Wonder, Meta has also begun developing larger and better performing AI models. At the same time, operational costs are being lowered to provide AI services to billions of users. The Muse Wonder is larger and has a higher operating cost than the existing small model 'Muse Spark'.
Accordingly, Meta is developing a specialized model that maintains Wonder's main performance while increasing operational efficiency. This is because in order to provide AI services to large-scale users, not only performance but also technology to reduce operating costs is important. Currently, Muse Wonder is provided through a "meta model API" that allows developers to utilize AI models and a coding tool called Muse Code.
The market also reacted positively. Meta shares rose up to 4.7% during the day after the release of the new model. Meta shares were up about 9% this year as of the closing price on the 9th. "You can use the model yourself and judge it," Wang said of how he doubted Meta had actually secured the highest level of AI technology.
Muse Wonder was first released as a developer service and will be applied to paid services of personal AI agent "Muse" in the future. Muse has not been officially released in Korea yet
