Meta releases lightweight model Muse Ember, featuring low-cost intelligent agent execution
The story so far
Meta Superintelligence Lab has released its smallest and most efficient AI model, Muse Ember, designed for the daily repetitive tasks of agents, including calling tools, handling small-scale coding tasks, routing requests, extracting data, and processing long documents. The model is available through Meta Model API and Muse Code, with an input price of 10 cents per million tokens and an output price of 50 cents per million tokens. Weights are not disclosed and can only be run on Meta's hosted infrastructure, with no local or quantized versions. Muse Ember is designed to be paired with the larger Muse Wonder: Muse Wonder is responsible for planning, and Ember handles a large amount of execution tasks. Reports say the agent can read screenshots, charts, and scanned documents, and can also handle voice memos, meeting recordings, and videos without needing another model. According to Artificial Analysis's review, Muse Ember scored 36 on the AI Index, equivalent to Claude Haiku 5.5, with a single-task cost of $0.14, two-thirds lower than the latter. Its input-output token prices are the same as Claude Haiku 5.5 and GPT-6 Luna, but with lower cached input costs. The same review shows it scored 68.1% on AutomationBench-AA, equivalent to Muse Wonder, but its knowledge accuracy is the lowest among the reviewed models, at 21.1%.
AI-written from the reports · Updated 8 days ago
Developments
2 developments
- 10 Nov, 02:00 · 2 reportsMeta releases Muse EmberTechCrunch · AI:Meta releases Muse Ember, a new lightweight model for agent execution
- 9 Nov, 08:00 · 4 reportsMuse Ember scores 36 on the AI IndexArtificial Analysis Articles:Meta's newest model Muse Ember scores 36 on the Artificial Analysis Intelligence Index, matching Claude Haiku 5.5 (max) at a third of the Cost per Task
Coverage timeline
Follow the coverage to see every side of the story.
- MarkTechPostMeta Releases Muse Ember, a 1M-Context Model for High-Volume Agent Work
Meta releases Muse Ember, its smallest and most efficient model, designed for repetitive steps in agents. Muse Ember scores 36 on the AA Intelligence Index, matching Claude Haiku 5.5 (max effort), with a cost of about 0.14 USD per task.
- ITHome · AIMeta launches lightweight AI model Muse Ember: benchmark scores comparable to Claude Haiku 5.5, with a single-task cost of just 1/3
Muse Ember is Meta's smallest and most efficient AI model, performing comparably to Claude Haiku 5.5 on the Artificial Analysis rankings, but with a single-task cost of only 1/3 and surpassing GPT-6 Luna.
- The DecoderMeta launches Muse Ember, a small AI model built for the repetitive work agents actually do
Meta releases Muse Ember, its smallest and most efficient AI model to date, designed for handling the repetitive tasks of AI agents, such as calling tools, processing small-scale coding tasks, routing requests, extracting data, and processing long documents.
- TechCrunch · AIFeaturedMeta releases Muse Ember, a new lightweight model for agent execution
Meta Superintelligence Lab releases lightweight model Muse Ember, priced at 10 cents per million tokens, focusing on high-frequency agent tasks such as tool calling, coding subtasks, routing, extraction, and long document parsing.
- Hacker News · 300+Muse Ember: Our smallest, most efficient model
Meta launches Muse Ember, its smallest and most efficient model designed for complex reasoning tasks, outperforming Claude Haiku 5.5 and GPT-6 Luna in smartness-to-cost ratio, with a cost of 0.10 USD per million input tokens and 0.50 USD per million output tokens, supporting a 1M context window and 512K output tokens.
- Artificial Analysis ArticlesMeta's newest model Muse Ember scores 36 on the Artificial Analysis Intelligence Index, matching Claude Haiku 5.5 (max) at a third of the Cost per Task
Muse Ember scores 36, comparable to Claude Haiku 5.5 but with a task cost of only 0.14 USD, two-thirds lower than the latter; its input/output token prices are the same as Claude Haiku 5.5 and GPT-6 Luna, but caching input costs are lower. It scored 68.1% on AutomationBench-AA, comparable to Muse Wonder, but has the lowest knowledge accuracy (21.1%)
Heat over time
- Now
- 3
- Peak
- 612 Nov, 06:00
- 24h change
- –
Hover or tap the chart to see each hour; use the arrow keys on a keyboard.