Meta releases Muse Ember, a new lightweight model for agent execution

Meta Superintelligence Lab launched Muse Ember on November 9, 2026, positioning it as the company's lightest and most cost-efficient model for agent execution work. The model, identified as muse-ember-1.0, features a 1 million token context window and a 512,000 token output ceiling. Meta prices fresh input at $0.10 per million tokens, cached input at $0.004 per million tokens, and output at $0.50 per million tokens. According to Meta's announcement, Muse Ember is "designed for complex reasoning execution and the work agents repeat all day: tool calls, coding subtasks, routing, extraction, and long document parsing." The model is available through Muse Code and the Meta Model API.

Meta emphasizes a tiered architecture where Muse Ember pairs with the larger Muse Wonder model. As Meta states, "Muse Wonder plans, Muse Ember handles the volume." In this pattern, Muse Wonder handles expensive but infrequent planning steps while Muse Ember executes high-volume repetitive tasks like tool calls, code edits, and document processing. Meta claims Muse Ember delivers more intelligence per dollar than Anthropic's Claude Haiku 5.5 or OpenAI's GPT-6 Luna, matching Haiku's best score at about a third of the cost and beating every Luna setting on Artificial Analysis benchmarks.

The model natively understands text, images, audio, and video, allowing agents to process screenshots, charts, scanned documents, voice memos, meeting recordings, and video without separate models. Meta evaluated Muse Ember under its Superintelligence Scaling Framework across chemical, biological, cybersecurity, and loss-of-control risk categories, finding it remained below high-risk thresholds. The model received the same core safety training as larger models, including evaluation of how it behaves when acting on users' behalf, such as respecting user intent, handling sensitive information, resisting hidden instructions, and seeking human input before high-impact actions.

Meta describes Muse Ember as the first specialized model trained using Muse Wonder Full, rounding out a portfolio that includes Muse Wonder for hardest problems, Muse Spark for coding and agentic workflows, and Muse Ember for repeatable tasks at scale at approximately $0.14 per Artificial Analysis task. The company states it has even more efficient models in training.