Meta says Muse Ember matches Haiku's top score at a third of the cost

Meta on Monday released Muse Ember, the smallest and most cost-efficient model from its Superintelligence Lab, aiming it at the repetitive work AI agents do all day, such as calling tools, handling coding subtasks and parsing long documents.

The release puts Meta in direct competition with the low-cost models Anthropic and OpenAI sell for the same work. Meta said Ember delivers more intelligence per dollar than Anthropic's Claude Haiku 5.5 or OpenAI's GPT-6 Luna, matching Haiku's best score on the Artificial Analysis benchmark at about a third of the cost.

At its highest reasoning setting, Ember costs about $0.14 per task on the Artificial Analysis Intelligence Index, compared with $0.41 for Haiku and $0.15 for Luna, The New Stack reported. It scores 36, the same as Haiku, while Luna scores 28.

Meta is charging $0.10 per million input tokens and $0.50 per million output tokens, the same standard rates as Luna and Haiku, with cached input at $0.004 per million tokens. The model has a 1 million-token context window and can produce up to 512K output tokens. It handles text, images, audio and video, which Meta said lets agents read screenshots and scanned documents or work from meeting recordings without a separate model.

Meta is pitching Ember as a worker for its larger models. As the company put it: “Muse Wonder plans, Muse Ember handles the volume.” Ember is the first model Meta trained using Muse Wonder Full, and the fourth released under the Muse brand.

Meta is keeping the model's weights private, as it did with earlier Muse releases, and Ember trails on at least one test: FrontierCode 2.0, which measures whether a repository maintainer would merge an agent's changes. The model is available now in Muse Code and through the Meta Model API as muse-ember-1.0.