Meta releases Muse Ember, a cheaper AI model for high volume tasks

In short: Meta has released Muse Ember, a lightweight AI model it says is built to handle lots of routine work at a low cost.

What happened

Meta Superintelligence Lab announced a new AI model called Muse Ember. Meta describes it as its lightest and most cost efficient model in the Muse family so far.

Meta says Ember is designed for work that AI “agents” do repeatedly all day. An AI agent is software that can take steps on your behalf, like calling other tools, pulling details out of documents, or doing small coding tasks (like a digital assistant that can also press buttons in other apps).

A key part of the announcement was price. Meta says Ember costs 10 cents per million input tokens. Tokens are small chunks of text that AI models read, like counting words in a very rough way. Meta also cited a composite benchmark from Artificial Analysis to argue Ember offers a strong balance of ability and cost compared with models like Anthropic’s Haiku and OpenAI’s Luna.

Meta says Ember is meant to work alongside a larger model such as Muse Wonder. In Meta’s example, a larger model can set the plan and rules, and Ember can apply them across many similar cases, like handling thousands of customer support messages.

Meta also says Ember can take in different kinds of media, including text, images, and audio. That means an agent could read screenshots or scanned documents, or use voice memos and meeting recordings, without needing a separate model for each format.

Why it matters

Many businesses want AI that is good enough and cheap enough to run at scale, especially for repetitive tasks. A lower-cost model could make it more realistic to use AI for high volume work, where every small cost adds up.

Source: TechCrunch AI