Announced 9 Nov 2026 · Sources checked
What is Muse Ember and what does it cost?
Meta Superintelligence Lab released Muse Ember on Monday, calling it the lab smallest and most cost-efficient AI model yet. Ember is designed for the work agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing.
Muse Ember is priced at $0.10 per million input tokens and $0.50 per million output tokens, with cached inputs at $0.004 per million tokens.
It offers a 1 million token context window and up to 512K output tokens, letting it take in a full codebase or document set in a single call.
According to Meta research blog, Ember is the fourth model released under the Muse brand, following the mid-weight Spark, Spark open-source twin Glimmer, and the Wonder frontier model.
It is the first trained using the Muse Wonder Full frontier model as a teacher.
How does it compare to competitors?
Meta says Ember delivers more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna at any reasoning setting. At its cheapest setting, Ember scores 33.9 on the Artificial Analysis Intelligence Index at $0.116 per task.
Claude Haiku 5.5 matches that score only at $0.248 per task — roughly double the cost.
On AutomationBench-AAE2E, a benchmark for business workflow automation, Ember at max reasoning scored 37.8 percent, compared to GPT-6 Luna at 20.2 percent and Claude Haiku 5.5 at 13.9 percent.
On Agents Last Exam, Ember scored 61.6 percent, ahead of GPT-6 Luna at 57.1 percent and Haiku 5.5 at 57.0 percent.
On OSWorld 2.1 for computer use, Ember achieved 69.5 percent (partial credit), slightly ahead of Haiku 5.5 at 68.9 percent and well ahead of Luna at 50.5 percent.
On PerceptionBench for fine-grained visual perception, Ember at max scored 80.6 percent, ahead of Haiku 5.5 at 78.6 percent and Luna at 62.8 percent.
What kind of tasks is it built for?
Meta designed Ember specifically for agent workflows where a model runs many small, structured steps.
In a customer-support agent, for instance, a heavier model like Muse Wonder would set rules for handling a particular issue, then Ember would apply those rules across tens of thousands of cases.
As Meta puts it: Muse Wonder plans, Muse Ember handles the volume. The model can carry a business workflow end to end, pulling records, calling tools in sequence, and updating systems across applications.
Muse Ember natively understands text, images, audio, and video, so agents can read screenshots, charts, and scanned documents, or work from voice memos, meeting recordings, and video without a separate model.
How does it fit into the Muse lineup?
Muse Ember rounds out the Meta model portfolio alongside the Wonder frontier model for the hardest problems, and Spark for coding and agentic workflows.
Ember is the budget option for repeatable coding and agentic tasks at scale, costing about $0.14 per Artificial Analysis task.
The Meta flagship consumer AI product, the Muse Agent, was released in September to what the company called unexpected success, with new features like Spaces and video chat rolling out earlier this month.
When Meta introduced Muse Wonder and Muse Wonder Full, it said it would use those models to develop specialized models available to everyone. Muse Ember is the first result of that effort.
Common questions
Is Muse Ember open source?
No. Meta has not announced open-source release plans for Muse Ember. It is available through the Meta Model API and Muse Code as muse-ember-1.0.
For comparison, Meta open-sourced the Muse Spark model as Muse Spark Open in August.
What safety testing did Meta run on Ember?
Meta evaluated Ember under its Superintelligence Scaling Framework, including chemical and biological, cybersecurity, and loss-of-control risks. The model remained below the framework high-risk thresholds across all assessed categories.
Sources & further reading
How this story was made
Written by Kristian Kostov with AI assistance and checked against the linked sources. Company performance claims are attributed to the company. Analysis reflects AiLookout’s interpretation; we have not independently tested the products discussed. The cover is an AI-generated editorial illustration, not a photograph or evidence of an actual event.
Our editorial standards
