Meta releases lightweight Muse Ember model for lower-cost AI agent workloads

Traders Union
All News
Business & Wealth
Companies
Meta Muse Ember lightweight AI model

Highlights

Meta launched Muse Ember, its most cost-efficient AI model, offering 10 cents per million input tokens and outperforming rivals like Anthropic's Haiku and OpenAI's Luna.
Muse Ember supports multimodal processing—including text, images, and audio—and can handle tasks such as parsing documents, reading screenshots, and processing video or audio inputs for enterprise workflows.
The Ember release expands Meta's Muse portfolio with a lower-cost execution layer as firms seek scalable agent deployments, following earlier launches like Muse Spark and consumer-oriented Muse Agent.

Cost positioning and model design

As reported by TechCrunch, citing Meta, Muse Ember is the company’s lightest and most cost-efficient AI model yet, built for complex reasoning execution in agent workflows that run repetitive tasks throughout the day. Meta says the model is designed to handle work such as tool calls, coding subtasks, routing, extraction and long document parsing, while pairing with larger models in broader agent systems.

Meta highlights Ember’s cost-per-task using Artificial Analysis’s composite benchmark score, presenting the model as both cheaper and smarter than rivals including Anthropic’s Haiku and OpenAI’s Luna. At 10 cents per million input tokens, Ember is also described as cost-competitive with open-source alternatives such as DeepSeek V4.1 Flash and GLM 5.3 Flash.

The company says Ember is intended to work alongside a heavier model such as Muse Wonder in an agentic setup. In a customer-support example, a larger model could set the rules for handling a case, while Ember applies those rules across tens of thousands of interactions, an approach Meta summarizes as, "Muse Wonder plans, Muse Ember handles the volume."

Broader product strategy and market impact

Meta also says Ember is trained to work across text, images and audio, widening the range of inputs it can process in enterprise and workflow settings. The company adds that agents using the model can read screenshots, charts and scanned documents, or work from voice memos, meeting recordings and video without requiring a separate model.

Ember is the fourth release under Meta’s Muse brand, following Spark, Spark’s open-source counterpart Glimmer, and the Wonder frontier model. The launch adds a lower-cost execution layer to Meta’s AI portfolio as companies look for ways to deploy agents at scale without sharply increasing inference spending.

Meta’s broader Muse push also includes the consumer-facing Muse Agent, which the company released in September and says saw unexpected success. New features including Spaces and video chat rolled out earlier this month, suggesting Meta is extending the Muse family across both enterprise automation and consumer AI products.

Earlier, we reported that a KPMG U.S. survey found only 9% of CEOs call AI agents transformational today.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.