According to Meta, Ember is designed to pair with a larger model such as Muse Wonder: the larger model plans and sets the rules, and Ember applies them at scale, for example across tens of thousands of customer-support cases. As the company puts it: "Muse Wonder plans, Muse Ember handles the volume." The model accepts text, images and audio, so agents can read screenshots, charts and scanned documents or work from voice memos and video without a separate model.
Meta presented Ember as cheaper and smarter than Anthropic's Haiku and OpenAI's Luna on Artificial Analysis's composite score, and according to TechCrunch it is also price-competitive with open models such as DeepSeek V4.1 Flash and GLM 5.3 Flash. These are the company's figures, and independent tests have not yet been published.
Ember is the fourth model in the Muse family, after Spark (April), its open-weights twin Glimmer (August) and the flagship Wonder (October).