Meta Unveils Lightweight Muse Ember for Efficient Agent Tasks

Meta Superintelligence Lab introduced a fresh AI model named Muse Ember this Monday, identifying it as the lab's lightest and most economical AI model to date. The company stated that Ember is "designed for complex reasoning execution and the work agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing."

Meta heavily highlighted Ember's cost-per-task efficiency, employing Artificial Analysis's composite benchmark score to demonstrate that Ember outperforms competitors such as Anthropic's Haiku and OpenAI's Luna in terms of affordability and intelligence.

With a price of 10 cents per million input tokens, Ember is notably cheaper than frontier rivals like Haiku and Luna. Furthermore, it remains cost-competitive with open-source alternatives including DeepSeek V4.1 Flash and GLM 5.3 Flash.

Focusing on scalability and low costs, Ember is engineered to function alongside a more substantial model like Meta's Muse Wonder within an agentic framework. For example, a heavier model might establish guidelines for addressing a specific customer-support issue, while a lighter model applies these rules across tens of thousands of similar cases. As Meta emphasizes in the announcement, "Muse Wonder plans, Muse Ember handles the volume."

Considering these high-volume applications, Meta also trained Ember to function effectively across various media types, such as text, images, and audio. This training results in a surprisingly flexible model capable of processing diverse inputs. As Meta notes, "agents can read screenshots, charts, and scanned documents, or work from voice memos, meeting recordings, and video without a separate model."

Ember represents the fourth model released under Meta's Muse brand, succeeding the mid-weight Spark (launched in April), Spark's open-source twin Glimmer (launched in August), and the Wonder frontier model (launched in October). The flagship consumer AI product, the Muse Agent, was released in September to unexpected success, with recent updates introducing features like Spaces and video chat earlier this month.