Meta launches Muse Ember for cheap agent work

Meta introduced Muse Ember, its smallest and most efficient Muse model, aimed at the repetitive work agents do all day, the company said on November 9, 2026. Meta lists the model on Muse Code and the Meta Model API as muse-ember-1.0 at $0.10 per million input tokens, $0.004 per million cached input tokens and $0.50 per million output tokens, with a 1M-token context window and up to 512K output tokens. TechCrunch and The Verge also covered the launch; the prices and context figures above are from Meta’s post.

Meta framed Ember as a volume workhorse for tool calls, coding subtasks, routing, extraction and long-document parsing. It says agents can take in a full codebase or document set in a single call, and that Ember natively handles text, images, audio and video. For harder problems, Meta’s own pairing advice is “Muse Wonder plans, Muse Ember handles the volume.” That is Meta’s product framing, not an independent workload study. Codema’s agent-harness explainer covers why the tooling around agents often matters as much as the model choice.

On cost, Meta claims Muse Ember “delivers more intelligence per dollar than Claude Haiku 5.5 or GPT-6 Luna at any reasoning setting,” matching Haiku’s best Artificial Analysis score at about a third of the cost and beating every Luna setting on Meta’s chart. Those comparisons are Meta’s reading of Artificial Analysis results dated November 9, 2026, using list pricing; they are not third-party audited unit economics. The price points land in the same budget tier builders already compare when they weigh API spend against self-hosting, as in Codema’s Mistral Large 4 self-hosting cost guide and LLM API pricing comparison. The practical pressure is on the cheap models agent builders already run for high-volume subtasks, not on frontier planners. Our AI models explained guide is the place to compare those cheaper models with the ChatGPT, Claude and Gemini flagships.