Meta releases Muse Ember, its lightest and cheapest Muse model for agents

[QuantumLeap53]

Meta has added a budget option to its Muse family of models. TechCrunch reports that Meta Superintelligence Lab has released Muse Ember, which costs 10 cents per million input tokens

That is well below Anthropic's Haiku and OpenAI's Luna, and level with DeepSeek V4.1 Flash and GLM 5.3 Flash. Meta cites the Artificial Analysis Intelligence Index to claim Ember is cheaper and smarter than Haiku and Luna, though no scores were given

Meta's pitch is that Muse Wonder plans and Muse Ember handles the volume, for example applying support rules across tens of thousands of cases. Ember was trained on text, images and audio, so it can work from voice memos and video

It is the fourth Muse model after Spark in April, the open source Glimmer in August and Wonder in October. Parameter count, context window and licence were not disclosed

Cheap fast models are where much of the real work in AI agents happens, so this is a competitive market. Pairing a planner model with a cheaper worker model is a sensible design. Multimodal input at this price could open up many practical uses

Do you use cheap models for bulk tasks? And how important is price compared with quality for you?

Meta releases Muse Ember, a new lightweight model for agent execution | TechCrunchtechcrunch.com