Meta launches Muse Wonder AI models for multistep agent tasks
Meta Platforms has released Muse Wonder and Muse Wonder Full, two large language models optimized for running AI agents that handle multistep personal and professional tasks. The models are positioned as Meta’s most capable yet for agentic workflows, such as managing complex scheduling or document drafting without human oversight.
The launch follows a month of rapid progress in AI agent infrastructure. When we covered Kore.ai’s Autoloop on 7 October, it was tuning live enterprise agents post-deployment; Iterate.ai’s Lifeboat, released three days earlier, promised higher GPU efficiency for concurrent agent sessions. AMD and Liquid AI have also targeted on-device execution, suggesting a market pivot from cloud-based scalability to hardware-aware, context-rich agents.
Meta’s announcement stops short of specifying deployment environments—whether cloud, edge, or on-device—but the emphasis on “longer, multistep jobs” aligns with the industry’s move toward persistent, task-oriented agents. The models’ capabilities and limitations remain untested outside Meta’s examples, leaving open how they compare to competitors like Microsoft’s real-time voice agents or Liquid AI’s privacy-focused architecture. Observers will watch for developer adoption and integration with existing agent platforms.
Sources: siliconangle.com
“Meta’s release signals the accelerating shift from single-turn chatbots to persistent, task-driven AI agents that operate across personal and professional workflows.”
Read the original reporting
The outlets below did the original reporting.
Related briefs
This brief was drafted automatically from the sources above and published under our editorial policy. Spotted an error? Tell us.