Meta Releases Muse Ember, a Low-Cost AI Model Designed to Handle High-Volume Agent Tasks

Meta has introduced Muse Ember, a lightweight artificial intelligence model designed to handle repetitive, high-volume tasks at lower cost, as the company expands its range of models for AI agents and enterprise applications.

Released on Monday by Meta Superintelligence Lab, Ember is positioned as the company's most cost-efficient model yet. It is designed for tasks that AI agents perform repeatedly, including tool calls, coding subtasks, information extraction, routing, and processing lengthy documents.

Meta is placing particular emphasis on the cost of completing tasks rather than model performance alone. Using Artificial Analysis's composite benchmark score, the company presented Ember as a lower-cost alternative to models including Anthropic's Haiku and OpenAI's Luna, while maintaining competitive intelligence performance.

The model costs $0.10 per million input tokens, a pricing level that puts it below the frontier competitors identified by Meta. The company also says Ember is cost-competitive with open-source alternatives such as DeepSeek V4.1 Flash and GLM 5.3 Flash.

The pricing, however, matters for businesses deploying AI agents across large volumes of customer interactions, documents and internal workflows. In these applications, a model may be called thousands or millions of times, making the cost of each request an important factor in determining whether automation delivers meaningful savings.

Meta's approach is to position Ember alongside a more powerful model rather than as a replacement for its larger systems. The company expects developers to combine the two within an agentic framework, allowing the more capable model to handle complex planning while Ember executes repetitive tasks.

As Meta describes the arrangement: "Muse Wonder plans, Muse Ember handles the volume."

The distinction between planning and execution is key to Ember's intended role.

For example, a customer-support agent could use Meta's larger Muse Wonder model to establish how a particular complaint should be handled, determine the relevant policies, and define the steps required to resolve it. Ember could then apply those instructions across tens of thousands of customer cases.

This division of work could reduce the need to use an expensive, more capable model for every interaction. It also offers businesses a way to manage the economics of AI deployment as they move from testing individual agents to operating them at scale.

A low price per token can improve the economics of automation, but businesses still need to account for accuracy, error correction, human oversight and the costs of running the wider agent infrastructure.

Meta has also trained Ember to process different types of information, including text, images and audio. This multimodal capability allows agents to work with screenshots, charts, scanned documents, voice recordings, meeting transcripts and video without requiring a separate model for each input type.

The company said the model can allow agents to "read screenshots, charts, and scanned documents, or work from voice memos, meeting recordings, and video without a separate model."

That flexibility could broaden the range of workflows businesses can automate. Customer-support systems, for example, could process written complaints alongside screenshots and voice messages. Administrative agents could extract information from scanned documents, while workplace tools could analyze meeting recordings and other media.

Ember is the fourth model released under Meta's Muse brand, following a series of launches that have expanded the company's AI development strategy.

The mid-weight Muse Spark was released in April, followed by Glimmer, an open-source counterpart to Spark, in August. Meta introduced its frontier Muse Wonder model in October, placing Ember at the smaller and more cost-sensitive end of the product range.

The launches indicate that Meta is developing a portfolio of models for different levels of complexity and deployment requirements rather than relying on a single system for every AI application.

The company's consumer-facing Muse Agent, launched in September, has also gained traction, according to the supplied report. Meta has since expanded the product with features including Spaces and video chat, which began rolling out earlier this month.

The broader strategy combines consumer-facing AI products with models intended for developers and businesses. While the Muse Agent provides a direct interface for users, Ember targets the underlying workload that could make autonomous and semi-autonomous agents more economical to operate.

The economics of AI agents remain an important challenge for the industry. More capable models can perform complex reasoning, but using them for every routine action can make large-scale deployment expensive. Smaller models offer a potential solution by handling predictable tasks at lower cost, provided they can maintain sufficient accuracy and reliability.

Meta's launch of Ember is believed to be an attempt to address that gap.