Meta Releases Muse Ember, a New Lightweight Agent Execution Model

Meta Superintelligence Lab has released Muse Ember, its lightest and most cost-efficient AI model to date. Ember is positioned for complex reasoning execution and the routine work agents do repeatedly: tool calls, coding subtasks, routing, extraction, and long document parsing. Meta highlighted cost-per-task, using Artificial Analysis’s composite benchmark score to show Ember as cheaper and smarter than competing models such as Anthropic’s Haiku and OpenAI’s Luna. At 10 cents per million input tokens, Ember is significantly cheaper than those frontier competitors and cost-competitive with open-source options like DeepSeek V4.1 Flash and GLM 5.3 Flash.

Ember is designed to work alongside a larger model in an agentic harness. Meta’s example is a customer-support agent: a heavier model establishes rules for handling an issue, then a lighter-weight model applies those rules across tens of thousands of cases. The announcement summarizes the division as “Muse Wonder plans, Muse Ember handles the volume.” To support volume-heavy use cases, Meta trained Ember to handle text, images, and audio, so agents can read screenshots, charts, scanned documents, or work from voice memos, meeting recordings, and video without needing a separate model.

Ember is the fourth model in Meta’s Muse line, following the mid-weight Spark, Spark’s open-source twin Glimmer, and the Wonder frontier model. The consumer-facing Muse Agent, released in September, has new features like Spaces and video chat rolling out earlier this month. The announcement gives Monday as the release date and does not include parameter counts or independent evaluation results beyond the referenced composite benchmark.

Meta releases Muse Ember, a new lightweight model for agent execution | TechCrunch

View Original