Meta releases Muse Ember, a lightweight model for high-volume agent execution
Reported by Mezha, Meta AI, TechCrunch and 4 more outlets7 sources ↓
Meta released Muse Ember for repeated tasks in AI agents, including tool calls, coding subtasks and document parsing. It costs 10 cents per million input tokens; Ember’s output price was not announced.
Key points
- Ember is designed to handle repeated agent tasks such as tool calls, routing and data extraction
- Meta lists input tokens at $0.10 and output tokens at $0.50 per million
- The Decoder reports a score of 36 for Ember versus 37 for Muse Spark 1.3, at $0.14 versus $2.39 per test task
- Ember is available through Meta Model API and Muse Code, according to The Decoder
Sources
- MezhaRead ↗Meta releases Muse Ember, a lightweight model for agent tasks
- Meta AIRead ↗Introducing Our Smallest, Most Efficient Model: Muse Ember
- TechCrunchRead ↗Meta releases Muse Ember, a new lightweight model for agent execution
- kobaran.comRead ↗Meta releases Muse Ember, a lightweight model for high-volume agent execution
- ecosistemastartupRead ↗Muse Ember: Meta lanza IA a US$0,10 por millón de tokens
Details
Meta Superintelligence Lab released Ember as a model for routine, repeated work in AI agents. The tasks named by Meta include tool calls, coding subtasks, routing, extraction and parsing long documents. Meta describes it as a model to pair with a larger system such as Muse Wonder, which can set rules that Ember then applies across many cases.
The Decoder reports that Ember cost $0.14 per task in an Artificial Analysis test, compared with $2.39 for Muse Spark 1.3; their scores were 36 and 37, respectively. It also reports that Anthropic’s Haiku scored 34.9% on FrontierCode 2.0-AA, compared with 27.2% for Ember. Meta used benchmark comparisons to make claims about cost and performance; the outlets do not independently establish those broader claims.
The Decoder says Ember is available through Meta Model API and Muse Code. It reports prices of $0.10 per million input tokens and $0.50 per million output tokens; the other articles give the input price but do not report the output price. The Decoder reports support for text, images, audio and video and a context window of up to one million tokens, while TechCrunch describes text, images and audio support.
Ember is the fourth model in Meta’s Muse lineup, following Spark, Glimmer and Wonder. TechCrunch reports Spark was released in April, Glimmer in August and Wonder in October.
Written from coverage as of
Images
- TechCrunch
- The Decoder
- ecosistemastartup
- TechCrunch
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh