What is the ‘Muse Wonder’ model, and can it help Meta stay ahead in AI race?

Meta has unveiled its most advanced frontier AI models, Muse Wonder and Muse Wonder-Full, designed to power personal and professional AI agents. The new models represent a significant leap over previous iterations, such as Muse Spark and Muse Glimmer, developed by Meta's Superintelligence Team. What sets Muse Wonder apart is its 'Active Memory' feature, which enables the model to retain user context for weeks or months, and 'Active Focus', allowing AI agents to persistently work on tasks until completion.

These capabilities position Muse Wonder as a potential game-changer in the AI race, particularly in personal and professional agentic work. Currently, access to Muse Wonder is limited to developers through Meta Model API and Muse Code, with pricing at $4 per million input tokens and $20 per million output tokens for enterprise users.

Muse Wonder Full, however, is exclusively available for Muse Code subscribers. The new models aim to consolidate Meta's position in the AI race following the success of its Muse agent, a significant turnaround for the social media giant after a period of AI-related struggles.