Meta launches Muse Ember to power low-cost AI agents

A new tier for AI agents

Meta Platforms Inc. has officially introduced Muse Ember, a new entry-level artificial intelligence model. The company designed this specific model to be highly cost-efficient, targeting users who need to power AI agents without the high expenses of frontier models. It is intended to handle the smaller, repetitive tasks that often make up the bulk of automated workflows.

The model represents a shift in how Meta approaches its AI lineup by focusing on affordability. While many companies prioritize the highest possible intelligence, Muse Ember is built to be a practical tool for high-volume operations. It serves as a bridge between simple automation and the complex reasoning capabilities of larger systems.

According to SiliconANGLE, Muse Ember is the first AI model the company trained using its flagship Muse Wonder algorithm. While Meta did not disclose the exact workflow, it is possible the flagship model was used for post-training. This process involves honing a model's capabilities by training it on additional data to refine its performance.

Beating the competition on price

Meta evaluated Muse Ember against its primary competitors using the Artificial Analysis Index. This index tracks nine different benchmarks, including specific use cases like financial analysis and coding. The results showed that Muse Ember matched the scores of Anthropic’s entry-level Claude Haiku 5.5 model but at about a third of the price.

The model also demonstrated superior performance in several other areas. When tested in 'max' mode—which uses the maximum amount of hardware available—Muse Ember outperformed OpenAI’s GPT-6 Luna algorithm across seven of the nine benchmarks. This suggests that the model remains competitive even when not strictly optimized for the lowest possible cost.

The most significant gap appeared in business task automation. Muse Ember clocked a 17% lead over its rivals on AutomationBench-AA. This specific benchmark consists of 657 tasks involving popular cloud applications like Salesforce and Gmail, highlighting the model's strength in corporate environments.

Technical specs and capabilities

Muse Ember is equipped with a context window of 1 million tokens. This allows the model to ingest and process very lengthy business documents in a single session. Users can also provide multimodal inputs, including images, audio, and video, to help the model understand complex requests.

The model is capable of generating responses up to 512,000 tokens in length. This high output capacity makes it suitable for generating long-form content or detailed reports based on the large amounts of data it can ingest. These features combined make it a versatile tool for various professional applications.

To further improve efficiency, Meta included a built-in cache for the model. This feature allows the system to reuse frequently used prompts to avoid processing the same information multiple times. When this cache is active, the cost for input tokens drops significantly to just 0.4 cents per million.

Integration into AI ensembles

The release of Muse Ember aligns with a growing trend in AI architecture known as ensembles. In these systems, a frontier model like Muse Wonder acts as a coordinator. It splits a user's request into smaller, manageable tasks and assigns them to lower-cost models like Muse Ember to execute.

By optimizing Muse Ember for this specific role, Meta is making it easier for developers to build scalable systems. This approach allows for

Community and industry insights

SiliconANGLE provides a platform for technology leaders to connect and share intelligence through the CUBE community. This network includes over 11,400 alumni who are shaping the future of industries like cloud computing, cybersecurity, and artificial intelligence. These leaders use the trusted-based network to create opportunities and stay informed on the latest technological shifts.

The outlet also offers a proprietary theCUBE AI Video Cloud to help companies make data-driven decisions. By leveraging a specific neural network, this tool helps organizations stay at the forefront of industry conversations. Readers can also support the mission of keeping content open and free by engaging with the community or using the Marketplace portal for AWS services.