Meta's Muse Ember undercuts Haiku 5.5 on cost
Meta's new efficient-tier Muse Ember matches Claude Haiku 5.5's Intelligence Index score of 36 at roughly a third of the cost per task, making it the cheapest model at that level.
Meta's Muse Ember, the newest model in its efficient tier, scores 36 on the Artificial Analysis Intelligence Index at max reasoning effort. Artificial Analysis found that it costs $0.14 per task, against $0.41 for Claude Haiku 5.5 (max) at the same score. That makes it the cheapest model scoring 36 or higher. At xhigh effort it scores 34 for $0.12.
Muse Ember costs the same per token as Claude Haiku 5.5 and GPT-6 Luna for input and output: $0.10 input and $0.50 output per million tokens. The cost gap comes from two things. Its cached input price is 60% lower at $0.004, and cached input makes up most of the cost in the agentic evaluations. It also uses roughly a third as many output tokens as Haiku 5.5. It is about 4x cheaper per task than DeepSeek V4.1 Flash 1017 (33, $0.60) and GLM-5.3-Flash (28, $0.58). It trails Meta's own Muse Wonder (max) by 8 points, at about 1/50th of the $6.95 cost per task.
The results vary by task. Muse Ember scores 68.1% on AutomationBench-AA, level with Muse Wonder. It beats Haiku 5.5 on Terminal-Bench 4.0 (40.9% vs 32.8%) but trails it on FrontierCode 2.0 (27.2% vs 34.9%). Running in Meta's Muse Code agent, it scores 46 on the Coding Agent Index, ahead of Haiku 5.5 in Claude Code (37) and GPT-6 Luna in Codex (31).
Knowledge is its weakest area. AA-Omniscience accuracy is 21.1%, mainly because it declines 64% of questions. The upside is a 17.4% hallucination rate, half that of Haiku 5.5.
The model is available through Meta's first-party API and Muse Code.