Meta ships Muse Ember, a small model built to run agent busywork
The fourth model in the Muse family prices input at $0.10 per million tokens and is meant to handle the repeat tool calls and coding subtasks a bigger planning model hands off.
Meta released Muse Ember, calling it the lightest and cheapest model it has shipped. It takes text, image, audio and video input, with a 1M-token context window and up to 512K tokens of output, and is live now through Meta’s Model API as muse-ember-1.0.
Pricing is $0.10 per million input tokens and $0.50 per million output, with cached input at $0.004 per million. Meta positions it as the bottom tier of a three-model stack: “Muse Wonder for the hardest problems, Muse Spark for coding and agentic workflows, and Muse Ember for repeatable coding and agentic tasks at scale.” Against Muse Spark 1.3, Meta says Ember scores close on its own Artificial Analysis Intelligence Index, 36 against 37, at roughly a seventeenth of the cost per task.
Why it matters: if you’re running an agent harness where a larger model plans and a cheaper one executes, repeated tool calls, routing, extraction, long document parsing, this is built for exactly that split. The 1M-token context and near-free cached input make it worth benchmarking against whatever small model you’re currently routing that work to.
The caveat: the benchmark comparison to Muse Spark comes from Meta’s own announcement, not an independent evaluation.