Meta's Muse Ember: a cheap small model for the work agents repeat all day
Agents run many small jobs: calling tools, writing bits of code, pulling facts out of long documents. Meta just released Muse Ember, its smallest and most efficient model, built for exactly that work. It got 300 points on Hacker News.
Price is the main pitch. Prices are counted in tokens, small pieces of text. TechCrunch says Ember is much cheaper than rivals like Claude Haiku and GPT Luna. Simon Willison notes that for shorter prompts it costs the same as GPT-6 Luna and Claude Haiku 5.5: $0.10 per million input tokens and $0.50 per million output tokens.
One Reddit user is not happy: they say Mark Zuckerberg promised to open source Muse Spark 1.2, but Meta keeps releasing closed models.
Cheap small models make agents cheaper to run, so we will see many more of them doing many more actions. That is useful, but every extra action is another chance to touch the wrong file or key. And the open-weight promise matters for who controls AI.
