THE CRUNCH
Meta Superintelligence Labs has released Muse Ember, which the lab describes as its lightest and most cost-efficient AI model yet. Ember is aimed at the repetitive work agents do all day: tool calls, coding subtasks, routing, extraction and long document parsing. Meta is leaning hard on price, positioning the model at 10 cents per million input tokens.
To make the case, Meta pointed to Artificial Analysis's composite benchmark score, presenting Ember as cheaper and smarter than Anthropic's Haiku and OpenAI's Luna, and cost-competitive with open-source options such as DeepSeek V4.1 Flash and GLM 5.3 Flash.
Ember is built to pair with a larger model, such as Meta's frontier Muse Wonder, in an agentic harness: the heavier model sets the rules for handling an issue, and Ember applies them across tens of thousands of cases. As Meta puts it, "Muse Wonder plans, Muse Ember handles the volume."
For volume-heavy work, Meta also trained Ember to handle text, images and audio, so agents can read screenshots, charts, scanned documents or work from voice memos and meeting recordings without a separate model. Ember is the fourth model under the Muse brand, following Spark, its open-source twin Glimmer and Wonder.
The launch fits Meta's broader Muse push. Wonder arrived in October, available to Muse Code subscribers and through the Meta Model API, with Meta saying it planned to power the Muse agent with the larger model over time. Ember now supplies the lightweight counterpart for that setup.