Meta releases Muse Ember, a new lightweight model for agent execution

What happened

On Monday, Meta Superintelligence Lab introduced Muse Ember, a new AI model designed for agent execution tasks. Muse Ember is Meta's lightest and most cost-efficient model to date, intended for complex reasoning and high-volume repetitive agent work. The model can process text, images, and audio, and is meant to be used alongside larger models like Muse Wonder for scalable workflows. Meta highlighted Ember's low cost per task and its ability to handle a wide range of input types.

Key details

• Muse Ember costs 10 cents per million input tokens.
• The model is positioned as cheaper and smarter than Anthropic’s Haiku and OpenAI’s Luna, according to Artificial Analysis’s benchmark.
• Ember is designed for tasks such as tool calls, coding subtasks, routing, extraction, and long document parsing.
• It can process multiple media types, including screenshots, charts, scanned documents, voice memos, meeting recordings, and video.
• Muse Ember is the fourth model in Meta’s Muse series, following Spark, Glimmer, and Wonder.

Why it matters

Companies and developers who need to automate large volumes of tasks, such as customer support or data extraction, can now do so at a lower cost using Muse Ember. The model’s ability to handle text, images, and audio in a single system reduces the need for multiple specialized models. This release allows organizations to scale their AI agent operations more affordably and flexibly, especially when combined with larger models for more complex planning.