Muse Ember

Muse Ember enters Meta's Muse family as a compact execution model rather than a general reasoning flagship. Meta describes it as its smallest and most efficient model, built for high-volume work and fast, agentic execution at scale. Its intended jobs are bounded and repetitive: coding subtasks, tool calls, and document summarization and extraction. The model accepts text, image, audio, and video as input and returns text, which makes it suited to pipelines that feed in mixed media — screenshots, clips, scanned documents, transcripts — and need short written outputs or structured actions back out.

Because the emphasis is throughput rather than depth, Meta's own guidance is to pair Muse Ember with the larger Muse Wonder or Muse Spark when a workflow needs deeper planning, reasoning, and orchestration. That framing suggests a division of labor: Muse Ember as the workhorse tier for fan-out steps, and the reasoning-oriented siblings carrying the harder parts of an agentic pipeline. Meta's launch material also framed it around per-task economics, presenting it as optimized for low cost on many small invocations, and shipped it inside Muse Code and the Meta Model API from day one, with availability across all Muse Code subscription plans at launch. Its practical fit is therefore teams running large volumes of straightforward subagent-style steps in code or document workflows.