Meta Unveils 10-Cent Muse Ember Model for Agent Tasks as It Targets High-Volume AI Work
Updated
Updated · TechCrunch · Nov 9
Meta Unveils 10-Cent Muse Ember Model for Agent Tasks as It Targets High-Volume AI Work
3 articles · Updated · TechCrunch · Nov 9
Summary
Meta’s Superintelligence Lab launched Muse Ember on Monday as its lightest and cheapest Muse model, built to execute repetitive agent work such as tool calls, coding subtasks, routing, extraction and long-document parsing.
At 10 cents per million input tokens, Ember is positioned on Artificial Analysis benchmarks as both cheaper and smarter than Anthropic’s Haiku and OpenAI’s Luna, while also competing on cost with open-source models including DeepSeek V4.1 Flash and GLM 5.3 Flash.
Meta said Ember is meant to pair with larger models such as Muse Wonder in an agentic setup—Wonder plans the workflow, while Ember applies those rules across high-volume tasks like customer-support cases.
The model also handles text, images and audio, letting agents work from screenshots, charts, scanned documents, voice memos, meeting recordings and video without a separate model.
Ember is the fourth Muse model after Spark, Glimmer and Wonder, extending a lineup that also includes the Muse Agent consumer product released in September.