According to CoinMeta, Meta has released its lightest model, AI Muse Ember, which is specifically designed to handle tasks such as repeated tool calls, code modifications, and document analysis by agents. Ember supports text, image, audio, and video inputs, with a context window of up to 1 million token. In the Intelligence Index of artificial intelligence analysis, Muse Ember and Claude Haiku both scored 36 points, but the cost per task was only 0.14 dollars, which is about one-third of the latter. Compared to Meta's own Muse Spark at 1.3, Ember also scored only 1 point lower, yet its cost dropped sharply to about 1/17. The key to the lower cost lies in token's efficiency. Ember and Haiku 5.5 have the same standard input and output unit prices, but Ember's cache read price is 60% lower, and the output token produced when completing the evaluation tasks is also only about one-third of Haiku's. Meta plans to let the large model Muse Wonder be responsible for planning complex tasks, while delegating a large amount of repetitive work to Ember.