On Monday, Meta’s Superintelligence Lab unveiled Muse Ember, a new AI model distinguished as the lab’s most lightweight and cost-effective solution to date. In its official announcement, Meta stated that Ember is “designed for complex reasoning execution and the routine tasks agents perform daily: tool calls, coding subtasks, routing, extraction, and comprehensive long document parsing.”
Meta particularly underscored Ember’s impressive cost-per-task efficiency, leveraging Artificial Analysis’s composite benchmark score to demonstrate Ember’s superior affordability and intelligence compared to competitors like Anthropic’s Haiku and OpenAI’s Luna.
At a rate of just 10 cents per million input tokens, Ember is considerably more economical than leading-edge competitors such as Haiku and Luna. Moreover, it achieves cost parity with open-source alternatives like DeepSeek V4.1 Flash and GLM 5.3 Flash.
Given the emphasis on low cost and high accuracy at scale, Ember is engineered to operate in conjunction with a more powerful model, like Meta’s Muse Wonder, within an agentic framework. For instance, in a customer-support scenario, a heavier model might establish guidelines for resolving a specific issue, then delegate a lighter-weight model to apply these rules across tens of thousands of cases. As Meta articulates in the announcement, “Muse Wonder plans, Muse Ember handles the volume.”
With high-volume use cases in mind, Meta also trained Ember to perform effectively across various media types, including text, images, and audio. This multi-modal capability makes the model remarkably versatile in the kinds of inputs it can process. As Meta explains, “agents can read screenshots, charts, and scanned documents, or work from voice memos, meeting recordings, and video without a separate model.”
Ember is the fourth model released under Meta’s Muse brand, following the mid-range Spark (launched in April), Spark’s open-source counterpart Glimmer (released in August), and the advanced Wonder frontier model (debuted in October). The flagship consumer AI product, the Muse Agent, was introduced in September and achieved unexpected success, with new features such as Spaces and video chat rolling out earlier this month.