Meta introduces Muse Ember, its smallest and most efficient model for high-volume tasks
Key Facts
Meta included the Muse Ember model in its developer news and described it as "our smallest and most efficient model." The company said it was designed for high-volume execution, with support for multimodal inputs and efficient handling of sub-agent coding, tool calling, document summarization and information extraction. This description positions Ember as a model for repetitive tasks that agents can carry out, while Meta’s messaging emphasizes size and efficiency. The available announcement did not specify a numerical operating cost or a quantitative comparison with other models. It also did not detail which types of multimodal inputs the model supports.
The mention of Ember comes as part of a broader effort to build models for Muse and Meta AI assistant. On November 6, Meta said it had begun rolling out specialized versions of its existing models, trained using Muse Wonder, to Muse and Meta AI users. This points to the company combining the development of specialized models with wider availability across its products, while Ember gives developers a model the company describes as suited to repetitive, high-volume execution. Meta had previously described the Muse Wonder series, introduced a few weeks before the November 6 announcement, as its most capable models at the time. It said they delivered advanced performance on agent-based personal and professional tasks.
In its announcement about the specialized models, Meta said models enhanced in Muse showed a 2.4-fold reduction in forgetting important information needed to help users achieve their goals. On the active memory test published by the company, the current model correctly used important context in 90% of tasks, compared with 76% for the previous model. Meta also reported a 67% improvement in completing everyday tasks involving unexpected problems. These results relate to Muse models as Meta presented them in its November 6 announcement, and provide context for the company’s focus on improving how agents handle context and obstacles while carrying out tasks. Meta did not attribute a separate figure to Muse Ember in these results.
Among the features of Muse Wonder highlighted by Meta were active memory and active focus. The company said active focus allows Muse to keep working toward a goal when it encounters unexpected challenges, linking the task-completion results it reported with the design of agents that can continue through workflows after a problem arises. This comes as Meta describes Ember as capable of efficiently handling tool calling, sub-agent coding, document summarization and information extraction. These tasks place the model on the execution side of the broader Muse ecosystem, which includes models for users and features that help the assistant stay on track with goals.
The case for expanding the use of assistants and agents also depends on a company’s ability to run them efficiently. According to comments by Mark Zuckerberg reported by the Associated Press during the third-quarter results announcement, more than 10 million people in the United States and Canada had created Muse accounts, and about 25% of them used it on a typical day. Zuckerberg said Muse requires substantial computing capacity to run, and that the company expects to improve cost efficiency by more than 10 times during the year following his remarks. These figures provide context on usage levels and the operational challenge facing Meta, but do not specify Ember’s cost or how much it alone will contribute to the expected efficiency gains.
The move comes as Meta recorded strong revenue growth alongside high capital spending to meet infrastructure needs. The company’s revenue rose 26% in the third quarter of 2026 to $64.60 billion, compared with $51.24 billion in the same period of 2025. At the same time, Meta raised the lower end of its 2026 capital expenditure forecast to $135 billion, from a previous estimate starting at $130 billion, while keeping the upper end at $145 billion. The figures reflect the financial context in which the company is expanding in AI, without the available sources linking Ember’s launch to a specific change in revenue or spending.
The share-price move reported in available coverage relates to the day Meta announced its third-quarter results, not the Ember launch session. Meta shares closed up 0.5% in regular trading on results day, then rose 2.9% in after-hours trading after the results were released. The company had said its specialized Muse models would roll out to Muse and Meta AI users, while the available information gave no separate timetable for expanding Ember’s availability or any financial impact specific to it. Meta’s announcement about Ember therefore remains a description of the model’s capabilities and intended uses, while the future cost-efficiency measure is tied to Zuckerberg’s forecast for the year following his comments about the third-quarter results.