Meta releases Muse Ember, a low-cost model for agent work
Priced at 10 cents per million input tokens, Ember targets high-volume agent tasks that need text, image and audio handling in one model.
Meta's Superintelligence Lab has launched Muse Ember, a compact model it describes as its most economical yet, aimed at repetitive agent work such as tool calls, coding subtasks, routing, extraction and parsing long documents. Meta prices it at 10 cents per million input tokens and argues that, on Artificial Analysis's composite benchmark, it beats Anthropic's Haiku and OpenAI's Luna on cost per task. For teams building agents, the pitch is a split workload: a heavier Muse Wonder sets the rules while Ember runs them at volume. Ember also accepts text, images and audio, so screenshots, scanned documents and recordings can be handled without a separate model. It is the fourth model in the Muse line, after Spark, its open-source twin Glimmer, and the Wonder frontier model.
Why it matters
The release targets teams running many repetitive agent steps, where per-call cost matters most. Meta's pitch is a split setup: the heavier Muse Wonder plans and Ember carries out the routine work, which could lower the cost of running agents at scale. Whether the cost claims hold up against Haiku and Luna is still something readers will need to test themselves.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.