Meta introduces Muse Ember: Focused on repetitive agent tasks

Meta announced its new AI model, Muse Ember, on November 9, 2026, and made it available through the Meta Model API and Muse Code. The company positions the model for frequently repeated agent tasks such as tool calling, small coding jobs, routing requests, extracting data, and processing long documents. The announcement did not specify an integration with any particular architecture or 3D software.

What tasks can Muse Ember handle?

According to Meta, Muse Ember can work with text, images, audio, and video. The model can also process up to one million tokens at a time. This multimodal input support is intended to handle different types of documents and recordings in a single model pipeline; however, the sources do not mention any functionality specifically for architectural drawings or 3D scenes.

Rather than serving as a standalone tool for large, complex creative work, the model is described as a way to carry out repetitive tasks within agent systems. According to TechCrunch, Meta highlights a division of labor in which the larger Muse Wonder model handles planning while Ember takes on high-volume execution steps.

What do the pricing and test results show?

Meta Model API pricing is listed at $0.10 per million input tokens and $0.50 per million output tokens. In the Artificial Analysis test reported by The Decoder, the cost per task was approximately $0.14. In the same comparison, Muse Spark 1.3 cost $2.39; their respective scores were reported as 36 and 37. These are results from a specific test; actual usage costs will depend on task length and the amount of input and output.

The picture is different for coding performance. In the FrontierCode 2.0-AA test reported by The Decoder, Muse Ember scored 27.2%, while Haiku reached 34.9%. Low cost therefore does not mean higher performance across every type of task. Although Meta says Ember led in most of its own tests, those results should be treated as the company’s assessment.

Impact on architecture and visualization workflows

For architecture firms, interior designers, and visualization teams, the model’s potential value lies in automating repetitive information tasks rather than directly modeling or rendering. Tasks such as extracting information from long documents, classifying requests, or routing workflows through tool calls could be handled as part of an agent. Support for image, audio, and video inputs could also make it possible to work with different media types in the same system; however, the sources do not explain whether it is compatible with project files or specific 3D applications.

Before evaluating the model, teams should take access methods, API usage, and token-based pricing into account. No information is provided about hardware requirements, licensing details, or direct compatibility with architecture software. Rather than treating Ember as a replacement for a production tool, a more realistic approach would be to test it on repetitive tasks in a controlled pilot.