Meta Releases Muse Ember, a Lightweight Multimodal Model for AI Agent Execution
Meta's new Muse Ember model targets agent execution tasks with multimodal input and a reported price of $0.10 per one million input tokens, with competitive benchmark claims flagged as vendor-side figures.
Meta has released Muse Ember, the fourth and lightest model in its Muse family, following Muse Spark, Muse Glimmer and Muse Wonder. The model is positioned for agent execution work, including tool calls, subtask coding, routing, data extraction and parsing of long documents. It is multimodal, handling text, images and audio, allowing agents to read screenshots, charts or scanned documents without a separate model.
According to TechCrunch, Muse Ember is priced at $0.10 per one million input tokens, competing with open-weight models in the same tier. Claims that the model is cheaper and more capable than Anthropic's Haiku and OpenAI's Luna are based on composite benchmark figures from Artificial Analysis cited in the report and should be treated as vendor-side benchmark claims.
Muse Ember is available through Meta's API, with no open-weight release announced at launch. Meta positions the model as a pairing with the larger Muse Wonder model, which handles planning while Ember processes high-volume execution tasks.