Driving the news: Meta on Monday unveiled Muse Wonder and Muse Wonder Full, models it says deliver industry-leading performance on a range of tasks involving personal and professional agents, memory and coding.
- The company released a series of benchmarks showing the models, previously known under the code name Watermelon, outperforming or matching the latest offerings from OpenAI and Anthropic.
- Muse Wonder is available through the Meta Model API and Muse Code, while the more capable Muse Wonder Full is available to Muse Code subscribers.
- Muse Wonder will also power future experiences in Meta’s Muse Assistant, initially for paid subscribers.
- The company says the models were trained using its one-gigawatt Prometheus data center.
What they say: “These are truly pioneering models, competitive with the best models in the world, and particularly at the forefront of personal and professional agent work,” Alexandr Wang, director of Meta Superintelligence Labs, told Axios.
Zoom: Meta emphasizes two capabilities: “Active Memory” and “Active Focus” are designed to help AI systems perform more useful work over extended periods of time.
- Active memory allows AI to retain important information about a user during conversations and retrieve it when it is relevant to a new task.
- Active Focus is designed to allow an AI agent to work toward a goal, including trying different approaches and adapting to new information when it encounters obstacles.
- Meta says the models also perform well in coding tasks and specialized professional work in fields such as law, finance and healthcare.
- “Memory is extremely important for personal agents, and that means it carries what matters to you across conversations for days, weeks, months, so that no one is re-explaining themselves every time,” says Wang.
Between the lines: Meta is presenting the new models as proof that it can compete with leading AI labs, rather than simply distributing AI technology through its massive consumer platforms.
- The company says the models perform particularly well on unfamiliar tasks, suggesting their capabilities extend beyond the benchmarks and tasks emphasized during training.
The big picture: Meta’s long-term ambition is what CEO Mark Zuckerberg called “personal superintelligence”: AI that understands an individual’s goals and preferences.
- Meta says the new models are a step towards this vision, with future specialist models earmarked for wider deployment across its products.
Reality check: Agents that remember personal information and act on a user’s behalf also introduce new privacy and security challenges.
Between the lines: Meta isn’t saying if it will open source any of the Wonder family models, but Wang noted that Meta recently open sourced the weights for Muse Glimmer as well as the technology needed for people to put their Muse Agent on hardware.
- “We are still committed to open source,” Wang said.
What to watch: Whether Meta’s claimed performance benefits hold up to independent testing – and how quickly the new technology reaches the billions of people using its apps.
- On the security front, Meta claims to have strengthened safeguards against cyber and biological abuse, as well as the protection of personal information. A model map with some details is expected to be released Monday, with a more comprehensive report to come.
Editor’s note: This story has been corrected to reflect that Meta’s open-weight model is Muse Glimmer (not Glitter).
Gn headline
