Meta✴ released Muse Ember, the lightest and most economical model of the Muse family in terms of token payment. The model is designed for mass execution of repetitive operations by AI agents and supports work with text, images and audio, including the processing and analysis of long documents.
Image source: Meta✴
Meta Lab Developers✴ Superintelligence Labs suggests using Ember in conjunction with the more powerful Muse Wonder. In this configuration, the older model will plan actions and determine the order of tasks, and the younger one will process a large number of requests. For example, Wonder will be able to develop support response rules that Ember will then apply to tens of thousands of tickets. Other use cases include programming and working with external tools.
According to the company, Muse Ember provides a superior cost-to-accuracy ratio than several competitors. For example, in the benchmark test Artificial Analysis Meta✴ compared it with the Anthropic Haiku and OpenAI Luna models, and set the cost of a million input tokens at just $0.10.
The model can also analyze screenshots, graphs, scanned documents, voice notes, meeting recordings and videos. Ember became the fourth development of the Muse family after Spark, Glimmer and Wonder. Muse Ember was presented today, November 9th.
If you notice an error, select it with the mouse and press CTRL+ENTER.
