Meta Releases Muse Ember, Lightweight Agent Execution Model

Meta Superintelligence Labs released Muse Ember on Monday, introducing its lightest and most cost-efficient AI model to date. The model is designed for complex reasoning execution and the repetitive workflows central to AI agents, including tool calls, coding subtasks, routing, information extraction and long-document parsing. Ember is positioned as a high-volume agent workhorse. At 10 cents per million input tokens, it is significantly cheaper than frontier competitors such as Anthropic’s Haiku and OpenAI’s Luna, while also being cost-competitive with open-source models including DeepSeek V4.1 Flash and GLM 5.3 Flash. Meta emphasized Ember’s cost-per-task performance, using Artificial Analysis’s composite benchmark score to argue that the model can be both cheaper and more capable than comparable alternatives. The model is intended to pair with a larger system such as Meta’s Muse Wonder within an agentic architecture. In that setup, a heavier model could define rules or strategy, while Ember applies those instructions across large numbers of tasks. Meta summarized the approach with the phrase “Muse Wonder plans, Muse Ember handles the volume.” Ember also reflects Meta’s push toward more flexible agent workflows. It was trained to operate across text, images and audio, allowing agents to process screenshots, charts, scanned documents, voice memos, meeting recordings and video without requiring separate specialized models. The release marks the fourth model under Meta’s Muse brand, following the mid-weight Spark in April, the open-source Glimmer in August, and the frontier Wonder model in October. It also extends the momentum behind Muse Agent, Meta’s flagship consumer AI product, which launched in September and has since rolled out features such as Spaces and video chat.