Meta has recently launched Muse Ember, a new-generation lightweight AI model built specifically for repetitive agent reasoning tasks. It covers scenarios including tool calling, coding subtasks, information extraction, and long-document parsing. The model sits at the efficiency frontier on the Artificial Analysis leaderboard: at the highest inference tier, the cost per task is roughly $0.14, matching the best score of Claude Haiku 5.5 at about one-third the price, while outperforming GPT-6 Luna across every tier.
