Meta AI launches Muse Ember, the smallest and most cost-efficient reasoning model
AI
Synthesized by Helena Ross, Strategic Think-Tank Desk
- Muse Ember costs $0.10 per million input tokens for prompts up to 200 k tokens.
- Pricing rises to $0.30 per million input tokens beyond 200 k, still below GPT-6 Luna and Claude Haiku at larger sizes.
- The model sits on the efficiency frontier, matching the Pareto line for intelligence-to-cost performance.
Meta AI’s Muse Ember offers low-price complex reasoning, positioning it against GPT-6 Luna and Claude Haiku.
«Introducing Our Smallest, Most Efficient Model: Muse Ember»
Summary and commentary by Aura-V. Facts, figures and opinions belong to the original.
Meta AI announced Muse Ember, its smallest model aimed at complex reasoning tasks such as tool calls, coding subtasks, routing, extraction, and long-document parsing. The launch details are posted on Simon Willison’s blog here. For prompts up to 200 k tokens, Muse Ember is priced at $0.10 per million input tokens, $0.004 for cached input, and $0.50 per million output tokens. Above that threshold the rates increase to $0.30 per million input, $0.012 cached, and $1.50 per million output, placing it between GPT-6 Luna and Claude Haiku in cost structure.
Compared with Claude Fable 5.2, which charges $8 per million input and $40 per million output, Muse Ember is dramatically cheaper. Meta highlights that the model lands on the “efficiency frontier” of the Artificial Analysis leaderboard, meaning no other model delivers higher intelligence per task cost.
The offering includes six reasoning levels—from minimal to max—though visual quality does not map cleanly to those levels. The “max” level output resembles results from Muse Wonder, a model priced 40 × higher.
Interested in implementing this technology inside your company?
Our studio designs custom agent meshes, data pipelines, and outbound GTM systems for forward-looking enterprises.
