Meta Platforms Inc. and its Superintelligence Lab released Muse Wonder and Muse Wonder Full on Monday. These state-of-the-art AI models are described as the company's most capable yet and represent Meta's first attempt at a frontier model.
The models are designed to run agents that handle professional and personal tasks, specifically focusing on longer, multistep jobs. The company's largest model rivals the world's best and includes new features intended to prevent these agents from going off the rails.
The release of Muse Wonder marks a departure from Meta's open-source Llama model, which had previously lagged behind the rest of the industry. These new models are intended to compete with cutting-edge AI offerings from rivals such as OpenAI and Anthropic.
This launch arrives as global downloads of Meta's personal-assistant agent have slowed. As a result, the Muse Wonder model will serve as a key test for the company's expensive and rebuilt superintelligence team.