Meta Superintelligence Lab released a new AI model on Monday called Muse Ember. This is the institute’s lightest and most cost-effective AI model to date. Meta said in a release that Ember is “designed to perform complex inferences, such as invoking tools, coding subtasks, routing, extracting, and parsing long documents, which work agents repeat throughout the day.”
Meta specifically highlighted Ember’s cost per task, using Artificial Analysis’s composite benchmark score to show that Ember is cheaper and smarter than competitors such as Anthropic’s Haiku and OpenAI’s Luna.
At 10 cents per million input tokens, Ember is significantly cheaper than cutting-edge competitors such as Haiku and Luna. Not only that, it’s cost-competitive against open source competitors such as DeepSeek V4.1 Flash and GLM 5.3 Flash.
Ember is designed to pair with larger models like Meta’s Muse Wonder in agent harnesses, with an emphasis on low cost and large-scale accuracy. For example, if you’re a customer support agent, a heavier model might establish rules for how to handle a particular issue, then send a lighter model to apply those rules to tens of thousands of cases. As Mehta said in a statement, “Muse Wonder will plan it, Muse Ember will handle the volume.”
With a large number of use cases in mind, Meta has trained Ember to work well with different types of media, including text, images, and audio. As a result, models are incredibly flexible in what kinds of inputs they can handle. As Meta says in the announcement, “Agents can read screenshots, charts, and scanned documents, and work from voice memos, meeting recordings, and videos without using a separate model.”
Ember is the fourth model under Meta’s Muse brand, following the mid-weight Spark (launched in April), Spark’s open source twin Glimmer (launched in August), and the Wonderfrontier model (launched in October). Its flagship consumer AI product, Muse Agent, was released in September to unexpected success, and new features like Spaces and video chat were rolled out earlier this month.
If you make a purchase through links in our articles, we may earn a small commission. This does not affect editorial independence.
