Back
AI Models9 November 20261 min read

Meta releases Muse Ember, its lightest and most cost-efficient model

Meta Superintelligence Labs introduced Muse Ember, its smallest and most efficient model, optimized for the high-volume repetitive work that agents perform all day: tool calls, coding subtasks, routing, extraction, and long-document parsing.

Meta releases Muse Ember, its lightest and most cost-efficient model

About the news

Key Details:

  • Pricing: $0.10 per million input tokens / $0.50 per million output tokens
  • Context: 1M-token window, up to 512K output tokens
  • Multimodal: text, image, audio, video
  • Performance: Matches Claude Haiku 5.5 on Artificial Analysis Intelligence Index at roughly one-third the cost per task; leads on AutomationBench agent workflows

Ember is the first model trained using Muse Wonder Full and is intended to pair with larger Muse models (Wonder for planning, Spark for coding) for efficient agent pipelines.

Sources