On November 9, 2026, Meta's Superintelligence Labs released Muse Ember, a lightweight model priced at just $0.10 per million input tokens — the cheapest model in the Muse family so far. Meta put it on the Meta Model API and into Muse Code at the same time, and the positioning is blunt: not the smartest, but the cheapest way to run the repetitive work agents do all day.
Where exactly is Ember cheaper?
Ember is the fourth model under the Muse brand, following Spark, released in April, Glimmer, open-sourced in August, and the flagship Wonder, released in October. In Meta's words, "Wonder plans, Ember handles the volume": Ember is built for tool calls, coding subtasks, routing, extraction and long-document parsing. On price, input costs $0.10 per million tokens and output $0.50, against $1.25 for input on its sibling Spark. Meta also cited Artificial Analysis composite benchmark scores, claiming Ember is cheaper and smarter than Anthropic's Haiku and OpenAI's Luna on a per-task cost basis. But no independent public benchmark scores or full technical specifications have appeared yet, so for now that claim is Meta's alone.
The point of cheap is the bill, not the specs
For developers, Ember's significance is not that it is "stronger", but that it runs volume more cheaply. An agent can call a model hundreds or thousands of times a day, and a price gap of an order of magnitude shows up directly in the monthly bill. With this move Meta also shifts the competition from "whose model is smartest" to "whose model delivers the best value", and lightweight models are becoming a tier every vendor has to fight for. But Ember only covers the execution layer — complex reasoning and planning still need larger models like Wonder, and using it as a general-purpose workhorse will expose its limits quickly.
Two things are worth watching next: whether independent tests from Artificial Analysis and others confirm Meta's "cheaper and smarter" claim, and how Ember actually performs inside Muse Code — for agent tool calls, the fear is not slowness but one mistake forcing the whole job to be redone. Meta has not yet published Ember's full technical specifications, so before choosing it, run a small comparison on your own real tasks rather than judging on price alone.
