Meta has expanded its artificial intelligence portfolio with the launch of Muse Ember, a compact AI model designed to handle repetitive, high-volume tasks at a lower cost. Meta CEO Mark Zuckerberg believes the new model delivers exceptional intelligence and performance compared with other AI models in its category.
The latest addition to Meta's Muse AI family is built to support AI agents that perform routine tasks continuously. These include coding assistance, information extraction, task routing and processing lengthy documents.
Zuckerberg announced the launch on X, describing Muse Ember as the company's smallest model that delivers higher intelligence per dollar than competing models in its category. The launch highlights Meta's growing focus on developing affordable and efficient AI systems for businesses and developers.
Muse Ember is a smaller AI model developed by Meta's Superintelligence Labs. Unlike advanced frontier AI models designed to handle highly complex reasoning and planning, Ember focuses on repetitive tasks that require speed, efficiency and lower operating costs.
Meta does not position Ember as a replacement for its larger models. Instead, it is designed to work alongside more powerful systems, including Muse Wonder.
Meta has introduced a collaborative approach in which larger AI models manage complex planning, while smaller models handle repetitive execution.
The company explains this approach through the concept: "Muse Wonder plans, Muse Ember handles the volume."
For example, in a customer support system, a larger AI model could establish guidelines for responding to customer complaints. Muse Ember could then apply those instructions across thousands of customer queries.
This arrangement allows businesses to distribute AI workloads more efficiently while potentially reducing operational expenses.
According to Alexandr Wang, head of Meta Superintelligence Labs, Muse Ember is the company's first publicly released model trained using Muse Wonder Full.
Muse Ember includes several capabilities that make it suitable for different business and technology applications.
The model natively supports multiple types of information, including:
Text
Images
Audio
Video
This means it can process screenshots, charts, scanned documents, voice recordings, meeting discussions and video content without requiring a separate model for each format.
Muse Ember offers a context window of one million tokens, allowing it to process substantial amounts of information within a single context.
It can also generate up to 512,000 output tokens, making it suitable for lengthy document processing and other large-scale tasks.
The model is particularly suited to token-intensive workloads performed by AI agents throughout the day.
Its potential applications include:
Breaking coding projects into smaller tasks.
Extracting information from large documents.
Routing requests to appropriate systems.
Processing repetitive business operations.
Handling large volumes of structured and unstructured information.
Meta is positioning Muse Ember against competing models, including Anthropic's Claude Haiku 5.5 and OpenAI's GPT-6 Luna.
According to benchmark results published by Meta, Ember outperformed these competitors in several evaluations.
One example is Terminal-Bench 4.0, which evaluates AI models on terminal-based tasks.
|
AI Model |
Terminal-Bench 4.0 Score |
|---|---|
|
Meta Muse Ember |
40.9% |
|
Anthropic Claude Haiku 5.5 |
39.2% |
|
OpenAI GPT-6 Luna |
16.4% |
Meta also highlighted results from the Artificial Analysis Intelligence Index. Ember and Claude Haiku 5.5 reportedly achieved rounded scores of 36 at their highest settings, while GPT-6 Luna scored 28.
These figures are based on Meta's published comparisons and should not be interpreted as independent confirmation of overall model superiority.
One of Muse Ember's major selling points is its relatively low operating cost.
Meta has announced the following pricing structure:
|
Usage Type |
Price |
|---|---|
|
Input tokens |
$0.10 per million tokens |
|
Output tokens |
$0.50 per million tokens |
Meta claims that Ember matches Claude Haiku 5.5's highest Artificial Analysis Intelligence Index score at approximately one-third of the cost. It also claims that the model outperforms every reasoning setting of OpenAI's GPT-6 Luna.
The pricing could make Muse Ember an attractive option for businesses and developers that need to process large volumes of AI requests without spending heavily on advanced models.
Alongside the Muse Ember launch, Meta has expanded the availability of its personal AI assistant to wearable devices.
Users can now access the Muse AI agent through compatible Apple Watch devices and smartwatches powered by Google's Wear OS.
The smartwatch version allows users to interact with the AI assistant through voice dictation or text.
It can help with everyday activities such as:
Coordinating schedules.
Collecting and summarising information.
Making reservations.
Managing routine requests.
The interface features a yeti-like avatar that users can tap to start interacting with the assistant.
Users need the latest Muse application installed on their smartphones to connect the assistant to a compatible smartwatch.
The supported software requirements include:
|
Device |
Minimum Software Requirement |
|---|---|
|
iPhone |
iOS 18.0 or later |
|
Apple Watch |
watchOS 9.0 or later |
|
Android smartphone |
Android 10 or later |
|
Wear OS smartwatch |
Compatible Wear OS application |
Android users can download the companion application through the Google Play Store.
Currently, Meta's Muse assistant is available in selected markets, including:
United States
Canada
Singapore
Japan
An India launch is reportedly being considered, although the company has not confirmed an official release date.
The launch of Muse Ember highlights an important development in the AI industry: smaller models are becoming increasingly important alongside advanced AI systems.
Rather than relying entirely on expensive, large-scale models, businesses can use a combination of powerful models for planning and smaller models for repetitive execution.
This approach could help organisations reduce AI operating costs, improve efficiency and automate everyday business activities.
With its multimodal capabilities, large context window and competitive pricing, Muse Ember represents Meta's effort to make AI agents more practical and affordable for large-scale applications.