Meta releases Muse Ember, a low-cost model for agent tasks

Meta says Muse Ember is designed for repeated agent work, including routing, extraction, coding subtasks and document parsing. The supplied coverage highlights its low token prices and multimodal inputs, while benchmark results and estimated task costs vary by evaluation and workload.

• Meta released Muse Ember as its lightest Muse model, intended for high-volume agent tasks such as tool calls, routing, extraction, coding subtasks and long-document parsing.· 2 sources
• Meta lists standard input at $0.10 per million tokens and says Ember accepts text, image, audio and video; one analysis estimates $0.14 per benchmark task, but actual costs depend on workload and token use.· 3 sources
• The articles describe pairing Ember with Meta’s larger Muse Wonder for planning or difficult decisions, while assigning Ember repeated execution; they also note weaker performance in some coding or factual-recall evaluations.· 3 sources