Meta launches Muse Ember model with 1 million token context for agent tasks
Also covered by 7 sources
What happened
Meta Superintelligence Labs introduced Muse Ember, a new compact AI model designed for repetitive agent tasks such as coding subtasks and document analysis. The model was launched on Meta’s hosted platforms and is not available for local deployment. Muse Ember can process up to 1 million tokens in a single context window and supports multiple input types, including text, images, audio, and video. The model is positioned for high-volume automation and is available through Meta’s Model API and Muse Code service.
Key details
- Muse Ember has a 1 million token context window and can output up to 512,000 tokens.
- It matches Claude Haiku 5.5 on the AA Intelligence Index with a score of 36.
- The cost per task is about $0.14, which is roughly one-third the cost of Claude Haiku 5.5 for similar tasks.
- Muse Ember’s weights are closed, meaning it can only be used via Meta’s hosted infrastructure.
- Early testing partners included Box, Cline, Oracle, Cloudflare, and You.com.
Why it matters
Businesses and developers using AI for automation and document processing can now access a model that handles large volumes of data at a lower cost per task compared to similar models. However, since Muse Ember is only available through Meta’s platforms and its model weights are not open, users cannot run it locally or customize it beyond Meta’s offerings. This setup may reduce costs for high-volume users but limits flexibility for those needing on-premise or customized deployments.