Meta released Muse Ember as a smaller multimodal model in its Muse family. The model targets agentic subtasks such as reading files, running tests, and targeted fixes while a larger model handles planning. It is available now under model ID muse-ember-1.0 in Muse Code and the Meta Model API with Responses API support.
Meta reports Ember pricing at $0.10 per million input tokens, $0.004 per million cached input tokens, and $0.50 per million output tokens. These rates match GPT-6 Luna and Claude Haiku 5.5 at standard short-context prices. The model supports a 1M-token context window and up to 512K output tokens.
On the Artificial Analysis Intelligence Index at max reasoning, Meta states Ember scores 36 at about $0.14 per task. This matches Haiku's rounded score at roughly one-third the cost and exceeds Luna's score for about a cent less per task. At the xhigh setting Ember scores about 34 at roughly $0.12 per task.
Meta keeps Ember model weights private. The company states Ember undercuts listed uncached input rates for DeepSeek V4.1 Flash 1017, Z.ai GLM-5.3 Flash, and MiniMax M3.1 Flash while matching or undercutting some output prices. On OSWorld 2.1 Ember completed 33.8 percent of tasks outright versus 27.8 percent for Haiku.