Meta releases Muse Ember, a new lightweight model for agent execution

Meta Superintelligence Lab released Muse Ember on Monday, according to TechCrunch. The company calls it the lab's lightest and most cost-efficient AI model so far. Meta says it was built for reasoning execution and the repetitive work AI agents do all day, such as tool calls, coding subtasks, routing, data extraction and parsing long documents.

Meta's pitch centers on cost per task. It cited Artificial Analysis's composite benchmark score to argue that Ember is both cheaper and more capable than Anthropic's Haiku and OpenAI's Luna. Ember is priced at 10 cents per million input tokens. TechCrunch notes that this undercuts those frontier rivals and is competitive even with open-source options like DeepSeek V4.1 Flash and GLM 5.3 Flash.

The model is meant to work alongside a larger one, such as Meta's Muse Wonder, inside an agent system. In a customer-support setup, for example, the heavier model might set the rules for handling an issue while Ember applies them across tens of thousands of cases. Meta summed up the split as "Muse Wonder plans, Muse Ember handles the volume." Ember also accepts text, images and audio. Meta says agents can use it to read screenshots, charts and scanned documents, or work from voice memos, meeting recordings and video, without needing a separate model.

Ember is the fourth model under the Muse brand. It follows the mid-sized Spark in April, its open-source counterpart Glimmer in August and the frontier model Wonder in October. Meta's consumer Muse Agent launched in September to unexpected success, TechCrunch reports, and gained features like Spaces and video chat earlier this month.