Meta has added another AI model to its Muse family. This time, it is Muse Ember, a smaller AI model designed to do repetitive work at a lower cost.

Meta CEO Mark Zuckerberg is confident that Ember is the most powerful model in its class.“Releasing Muse Ember today — our smallest model that delivers higher intelligence per dollar than others in its class,” Mark Zuckerberg wrote on X. The Meta chief claimed that Ember was built for “token-heavy work” done by AI agents all day long.

This includes, “coding subtasks, routing, extraction, and long document parsing.” Mark Zuckerberg shared on X. What is Muse Ember?Ember is not a frontier AI model.

Rather, Meta says, it is designed to work alongside larger models such as Muse Wonder. A larger model can plan a complicated assignment before handing repetitive steps to Ember.

Meta described the arrangement by saying, “Muse Wonder plans, Muse Ember handles the volume.” In a customer-support system, for example, a larger model could establish rules for handling an issue while Ember applies them across thousands of cases.According to Alexandr Wang, who leads Meta’s Superintelligence Labs, Muse Ember was the company’s first released model trained using Muse Wonder Full. Muse Ember natively supports text, images, audio and video.

It can process screenshots, charts, scanned documents, voice memos, meeting recordings and videos without requiring a separate model. It has a context window of one million tokens and can generate up to 512,000 output tokens.Ember takes on Anthropic’s Claude Haiku 5.5 and OpenAI’s GPT-6 Luna models.

In terms of benchmarks shared by Meta, Ember outperforms the two models in most cases. For instance, the model scored 40.9 per cent on Terminal-Bench 4.0, compared with 39.2 per cent for Haiku and 16.4 per cent for Luna.

Muse Ember costs $0.10 per million input tokens, and $0.50 per million output tokens. Meta said Ember matched the highest Artificial Analysis Intelligence Index score of Anthropic’s Claude Haiku 5.5 at about a third of the cost and outperformed every reasoning setting of OpenAI’s GPT-6 Luna.

The published results gave Ember and Haiku rounded scores of 36 at their highest settings, while Luna scored 28.Meta’s Muse now runs on your smartwatchSeparately, Meta has expanded its Muse personal AI agent to smartwatches. The company announced that you can run the AI agent on your Apple Watch or smartwatches powered by Google’s Wear OS.

Muse's official X account posted the update. You can access Muse on the smartwatch.

You can use the AI for dictation or text to request tasks such as coordinating schedules, gathering information and making reservations. The interface displays Muse’s yeti-like avatar, which users can tap to begin an interaction.To connect Muse to your smartwatch, you will need the latest Muse app on your smartphone.

The iPhone app requires iOS 18.0 or later, while the Apple Watch companion requires watchOS 9.0 or later. The Android app supports Android 10 and later, and you can download the Wear OS app through the Google Play Store.

Though the Muse assistant remains available in select markets – the US, Canada, Singapore and Japan. An India release is rumoured to be on the cards.- Ends