On November 9, Meta officially launched Muse Ember, its smallest and most efficient high-performance AI model to date. The model is purpose-built for the routine, repetitive tasks that AI agents perform every day, covering tool calling, small-scale coding, request routing, data extraction, and long-document processing.
In terms of core performance and cost, Muse Ember demonstrates strong market competitiveness. According to the Artificial Analysis Intelligence Index (AII), it outperforms Claude Haiku 5.5 and GPT-6 Luna at every inference level.
On the AII benchmark, Muse Ember scored 36 points, nearly matching the larger Muse Spark 1.3 (37 points), but its per-task processing cost was reduced to about $0.14, far below Muse Spark 1.3's $2.39. However, in the real-world coding test FrontierCode 2.0-AA, Haiku performed better, scoring 34.9%, while Muse Ember scored 27.2%.
In terms of multimodality and input specifications, the model fully supports text, image, audio, and video processing, with a single context window of up to one million tokens. Muse Ember is now officially available through the Meta Model API and Muse Code, with input and output token pricing set at $0.10 and $0.50 per million tokens, respectively.
This development marks a shift in the AI industry from chasing raw parameter scale to improving inference economics. Lightweight, cost-effective, agent-specific models may accelerate the large-scale deployment of AI agents in vertical business scenarios.
