Tuesday, 10 November 2026

Meta’s Muse Ember is a $0.10 model built for agent busywork

Meta's new Muse Ember is its cheapest Muse model so far: 10 cents per million tokens in, aimed at repetitive agent tasks.

Key points

  • Muse Ember costs $0.10 per million input tokens and $0.50 per million output tokens.
  • It is available in Muse Code and the Meta Model API as muse-ember-1.0.
  • Meta's benchmark claims are its own; Ember trails Haiku on FrontierCode 2.0.

Meta has released Muse Ember, the smallest and cheapest model in its Muse family. Meta charges 10 cents for every million tokens it reads and 50 cents for every million it writes, and aims it at repetitive agent chores such as calling tools, small coding jobs, routing requests, pulling out data and going through long documents.

The model ID is muse-ember-1.0, and developers reach it from Muse Code or Meta’s own Model API. Meta is keeping the model weights private, as it has with earlier Muse releases. Ember takes text, images, audio and video, has a 1 million token context window and can produce up to 512,000 tokens of output.

Meta pitches it as the workhorse beside the bigger Muse Wonder: Wonder plans, Ember handles the volume. A customer-support agent could let the larger model write the rules for a complaint type and then hand thousands of cases to Ember. Meta says Ember is the first model it trained using Muse Wonder Full, and that it can operate desktop apps from screenshots.

Rivals charge about the same. The New Stack’s side-by-side check found the same short-prompt rates on small models from OpenAI and Anthropic, and higher input prices on three Chinese alternatives, among them DeepSeek’s.

The performance claims are Meta’s own and lean on one test. On the Artificial Analysis Intelligence Index, Meta puts Ember at about 36, the same rounded score as Haiku, at roughly $0.14 per task, where Haiku costs nearly three times as much; Luna is close on cost but scores 28. Meta also reports a lead on its own Personal Superintelligence Index. On FrontierCode 2.0, which asks whether a maintainer would merge an agent’s code, Ember scores 27.2%, behind Haiku at 34.9% and DeepSeek at 28.5%.

Ember is the fourth model under the Muse name, after Spark in April, the open Glimmer in August and Wonder in October. The Muse agent app arrived in September, and Spaces was added earlier this month. On the OSWorld 2.1 test of computer use, Ember scored 69.5% for partial credit against Haiku’s 68.9%, and finished 33.8% of tasks outright against 27.8% for Haiku. Meta has not said when Ember will reach the consumer app, and the announcement gives no timeline for a consumer rollout.

TPRO Team

The Tech Prolonged newsroom desk. Stories under this byline cover AI, the web and general tech, and are reviewed by our editors before publishing.