Meta Releases Muse Ember, a Low-Cost Model Built for Repetitive Agent Work

Meta said Ember was "designed for complex reasoning execution and the work agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing." The model is available through Meta's Model API at $0.10 per million input tokens and $0.50 per million output tokens.

That input price undercuts small models from the leading labs, including Anthropic's Haiku and OpenAI's Luna, and puts Ember on a par with low-cost open-weight options such as DeepSeek V4.1 Flash and GLM 5.3 Flash, according to TechCrunch. For companies running agents, which can make thousands of model calls to finish a single job, the cost of each completed task increasingly decides which model gets used. Meta built its launch pitch around that measure, using Artificial Analysis's composite benchmark score to present Ember as both cheaper and more capable than Haiku and Luna.

Those comparisons come from Meta's own presentation. Crypto Briefing reported that no independent benchmarks or detailed technical specifications had been published at launch, leaving open how Ember performs against Meta's own Spark and Wonder models or against rival small models.

Meta is positioning Ember as a partner to a larger model in an agent system rather than as a standalone product. In a customer support setting, for example, a heavier model could set the rules for handling a type of complaint and Ember could then apply those rules across tens of thousands of cases. "Muse Wonder plans, Muse Ember handles the volume," Meta said in its announcement.

The model also accepts several kinds of input, including text, images and audio, so that a single model can cover most of an agent's routine work. Meta said "agents can read screenshots, charts, and scanned documents, or work from voice memos, meeting recordings, and video without a separate model."

The pricing is a steep step down from the rest of the lineup. Muse Spark 1.3 costs about $1.25 per million input tokens and $4.25 per million output tokens, roughly 12.5 times Ember's input price and about 8.5 times its output price, according to Crypto Briefing. The publication also noted that Ember's list rates closely match Meta's discounted Contributor tier, under which developers allow Meta to use their interaction data to train future models.

Ember is the fourth model released under the Muse brand since Meta reorganised its AI work under Meta Superintelligence Labs. It follows the mid-sized Spark in April, Glimmer, an open-source counterpart to Spark, in August, and the frontier model Wonder in October. Meta's consumer product, the Muse Agent, launched in September, and features including Spaces, which lets users build mini apps by describing them, and video chat were added earlier this month.