At ten cents per million input tokens, Meta wants to capture the routine work of AI agents
Muse Ember is the smallest model Meta has released so far and is meant mainly to handle repetitive jobs such as extracting data and calling tools at low cost.
Meta on Monday unveiled Muse Ember, a compact language model for simple but numerous tasks carried out by AI agents. Digital Trends lists, among other things, calling tools, small coding jobs, routing requests and parsing long documents. The model handles text, images, audio and video and can take in up to a million tokens at once. Mark Zuckerberg announced the launch himself on X.
The price is 0.10 dollars per million input tokens and 0.50 dollars for output. According to Interesting Engineering, which relies on the comparison site Artificial Analysis, a task with Muse Ember costs about 14 cents on average, against roughly 41 cents for Claude Haiku 5.5 at a similar score. Meta's larger Muse Spark 1.3 scores barely one point higher but costs 2.39 dollars per task.
In Meta's own tests, Ember posts clearly better results on automation tasks than OpenAI's GPT-6 Luna and Claude Haiku 5.5, while Haiku stays ahead on real-world coding jobs. Both sites stress that these figures come from Meta itself. The model is available through the Meta Model API, Muse Code, Google Cloud and Oracle Cloud. Whether the weights will be released is not known.