Meta releases Muse Ember, a new lightweight model for agent execution
Summary
- Meta Superintelligence Lab has released Muse Ember, its lightest and most cost-efficient AI model designed for agent execution and complex reasoning.
- Ember is priced at 10 cents per million input tokens, positioning it competitively against open-source flash models and smaller frontier alternatives.
- The model is optimized to handle repetitive agent tasks including tool calls, coding subtasks, routing, extraction, and long document parsing.
- Ember supports multimodal inputs across text, images, and audio, allowing agents to process screenshots, charts, voice memos, and video without separate models.
- Meta intends for Ember to pair with larger models like Muse Wonder, where the frontier model plans and Ember handles high-volume execution.
Why it matters
The release of cost-efficient, multimodal lightweight models like Muse Ember lowers the economic barrier for deploying high-volume AI agents, enabling organizations to scale automated workflows across diverse media types.