The new model is priced at $0.10 per million input tokens and is designed to handle repetitive AI agent tasks, including coding, document processing and tool execution.
Meta has introduced Muse Ember, a lightweight artificial intelligence model designed to support high-volume tasks performed by AI agents. Developed by Meta Superintelligence Labs, the model focuses on reducing the cost of executing repetitive operations while maintaining capabilities across reasoning, coding and multimodal processing.
Announced on November 9, 2026, Muse Ember is positioned as Meta's lightest and most cost-efficient model in its Muse family. It is designed to handle tasks such as tool calls, coding subtasks, routing, information extraction and processing lengthy documents.
Unlike larger AI models that are used for complex planning and advanced reasoning, Muse Ember is intended to manage routine execution tasks that AI agents perform repeatedly across enterprise workflows.
Meta has priced the model at $0.10 per million input tokens, placing it among lower-cost options for organisations deploying AI applications at scale.
According to the company, Ember offers competitive performance relative to models including Anthropic's Haiku and OpenAI's Luna, based on comparisons using Artificial Analysis benchmarks. Meta also positions its pricing against open-source alternatives such as DeepSeek V4.1 Flash and GLM 5.3 Flash.
The model is designed to operate alongside more advanced systems, including Meta's Muse Wonder. Under this approach, a larger model can manage planning and decision-making, while Ember executes individual tasks across large volumes of requests.
For example, in customer service operations, a larger AI model could establish instructions for resolving customer queries, while Ember processes thousands of individual cases according to those instructions.
This division of responsibilities is intended to help organisations manage the computing costs associated with deploying AI agents across business processes.
Muse Ember also supports multimodal inputs, allowing it to process text, images and audio. The model can analyse screenshots, charts, scanned documents, voice recordings and video-based information without requiring separate models for each input type.
These capabilities could support enterprise applications involving document processing, customer interactions, information retrieval and workflow automation.
Muse Ember is the fourth model introduced under Meta's Muse portfolio. It follows Muse Spark, launched in April 2026, the open-source Muse Glimmer released in August, and Muse Wonder, introduced in October.
Meta also launched its consumer-facing Muse Agent in September, expanding its AI offerings beyond individual models into agent-based applications.
The introduction of Ember reflects Meta's approach to developing models for different stages of AI workflows, rather than relying exclusively on larger systems for every task.
The company has not disclosed detailed enterprise adoption figures or independent production performance results for the newly released model.
With Muse Ember, Meta is expanding its AI portfolio to address the cost and operational requirements of organisations deploying agentic AI across high-volume applications.
Disclaimer: This article may include information derived from interviews, press releases, public statements, research, company communications and other publicly available or third-party sources. Such material may be summarised, paraphrased or contextualised for journalistic and editorial purposes. All rights in third-party content remain with their respective owners.