Particle.news

Meta Launches Muse Ember as a Low‑Cost Reasoning Model

The model is priced to serve high-volume agent workloads with a Contributor option that gives Meta rights to use interaction data for training.

Overview

  • Meta released Muse Ember on Monday, November 9, 2026, positioning it as the Muse family’s lightest, most cost‑efficient reasoning model for developers.
  • Ember’s list prices are $0.10 per million input tokens and $0.50 per million output tokens, rates aimed at making routine agent tasks cheaper to run at scale.
  • Meta says Ember supports multimodal inputs — text, images, audio, and video — and is designed for repeated agent work such as tool calls, routing, extraction, coding subtasks, and long‑document parsing.
  • The lowest pricing closely matches Meta’s discounted Contributor tier, which lowers costs in exchange for permission to use customers’ interaction data to improve future models, a tradeoff that raises privacy and compliance questions for regulated users.
  • Independent, public benchmarks and full technical disclosures for Ember remain limited even as Meta cites composite vendor benchmarks showing cost‑performance advantages, leaving enterprises to weigh savings against uncertain real‑world performance.