Meta Launches Muse Ember AI Model for Efficient Agent Tasks
Meta has introduced Muse Ember, a new AI model built for developers and businesses that rely on AI agents. The model focuses on smaller jobs inside larger workflows. These include reading files, running code tests, extracting information, and calling external tools.
Meta says Muse Ember can match the performance of larger models while costing less. Therefore, it suits tasks that agents need to repeat many times. The company has now made the model available through its developer platforms.
Muse Ember supports a 1-million-token context window. It can generate up to 512,000 output tokens. The model accepts text, images, audio, and video as inputs. As a result, developers can analyse long documents, review screenshots, and work with large codebases without splitting them into smaller requests.
Meta shared benchmark results comparing Muse Ember with Anthropic’s Claude Haiku 5.5 and OpenAI’s GPT-6 Luna. In the Artificial Analysis Intelligence Index, the model scored 36 points. This matched Claude Haiku 5.5 and exceeded GPT-6 Luna’s score of 28 points. Meta also claims Muse Ember delivers performance similar to Claude Haiku 5.5 at roughly one-third of the cost.
In AutomationBench-AA, which tests automated task handling, Muse Ember scored 37.8 percent. This was higher than GPT-6 Luna’s 20.2 percent and Claude Haiku 5.5’s 13.9 percent. It also scored 40.9 percent in Terminal-Bench 4.0, slightly ahead of Claude’s 39.2 percent. However, Claude’s model performed better in FrontierCode 2.0-AA, scoring 34.9 percent against Muse Ember’s 27.2 percent.
Muse Ember is the first model Meta says it trained using Muse Wonder Full. The company also ran safety evaluations covering cybersecurity threats, harmful instructions, and other risks. Meta said the model did not cross any of its high-risk thresholds.
Developers can access Muse Ember through Meta’s Model API and Muse Code under the name muse-ember-1.0. Pricing stands at $0.10 (roughly Rs. 9.36) per million input tokens and $0.50 (roughly Rs. 47) per million output tokens. Cached input tokens cost $0.004 (roughly Rs. 0.37) per million. Meta has also added support for the Responses API and made the model available on Google Cloud and Oracle Cloud.