Meta Introduces Muse Ember Model

Meta has introduced Muse Ember, described by the company as its smallest and most efficient model designed for complex reasoning execution and tasks agents repeat all day, such as tool calls, coding subtasks, routing, extraction, and long document parsing.

The model is available on the Meta Model API, priced at $0.10 per million input tokens and $0.50 per million output tokens. According to Meta, Muse Ember delivers more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna at any reasoning setting, matching Haiku's highest score at about a third of the cost and beating every Luna setting. Meta also states the model sits on the efficiency frontier of the Artificial Analysis leaderboard, with no model scoring higher for less per task. Estimated costs per task for Muse Ember sit at $0.116 for a score of 33.9 and $0.136 for a score of 35.8, compared to figures listed for Claude Haiku 5.5 ranging from $0.092 to $0.411 and GPT-6 Luna from $0.073 to $0.152.

Muse Ember features a 1M-token context window and supports up to 512K output tokens. Meta states the model natively understands text, images, audio, and video. Evaluation results provided by Meta show Muse Ember scoring 37.8 on AutomationBench-AA E2E business workflows, 61.6 on Agents' Last Exam, 46.8 on the Personal Superintelligence Index, 40.9 on Terminal-Bench 4.0, 46 on the Coding Agent Index-AA, 27.2 on FrontierCode 2.0-AA, 22.8 on GDP.pdf-AA, 69.5 (partial) and 33.8 (binary) on OSWorld 2.1, and 80.6 on PerceptionBench with tools.

Meta suggests pairing Muse Ember with Muse Wonder for complex problems, stating Muse Wonder handles planning while Muse Ember handles volume.

Before release, the model was evaluated under the Meta Superintelligence Scaling Framework for chemical and biological, cybersecurity, and loss-of-control risks. Meta reports that Muse Ember remained below high-risk thresholds while receiving the same core safety training and testing as larger models.