On Monday, Meta Superintelligence Lab launched Muse Ember, marking the organization's lightest and most cost-efficient AI model designed specifically for agent execution. According to TechCrunch, the new system handles repetitive tasks like tool calls, coding subtasks, and long document parsing while maintaining a low price point.
Key takeaways
- Meta Superintelligence Lab released Muse Ember on Monday.
- Ember costs 10 cents per million input tokens.
- The model handles text, images, and audio inputs natively.
- It is the fourth model released under the Muse brand.
What is Muse Ember?
TechCrunch reports that Meta designed the lightweight model for complex reasoning execution and repetitive agent work. The system tackles routing, extraction, and document parsing. It aims to deliver high accuracy at scale without driving up operational expenses.
Meta highlighted the cost-per-task metric by using Artificial Analysis data. The benchmarks show that Ember beats competitors such as Anthropic's Haiku and OpenAI's Luna on both price and smarts. It also competes directly with open-source options like DeepSeek V4.1 Flash and GLM 5.3 Flash at just 10 cents per million input tokens.
How does the agent execution model work with other systems?
Developers can pair the new agent execution model with a larger frontier system like Meta's Muse Wonder. In customer support setups, a heavier model sets the rules, and the lighter system applies them across thousands of cases. Meta describes this division of labor by stating that Muse Wonder plans while Muse Ember handles the volume.
Also, the system processes text, images, and audio without requiring separate models. Agents can read screenshots, charts, and scanned documents, or work from voice memos, meeting recordings, and video.
What happens next?
Ember becomes the fourth model under the Muse brand. It follows Spark in April, Glimmer in August, and Wonder in October. The Muse Agent launched in September, and Meta continues expanding the product ecosystem with new video chat features and Spaces.
Frequently asked questions
Muse Ember is a lightweight and cost-efficient AI model released by Meta Superintelligence Lab on Monday. It is built for agent execution, reasoning, tool calls, coding subtasks, and document parsing.
According to TechCrunch, Muse Ember costs 10 cents per million input tokens, making it cheaper than competitors like Anthropic's Haiku and OpenAI's Luna.
Muse Ember processes text, images, and audio inputs. This allows agents to work with screenshots, scanned documents, voice memos, meeting recordings, and video.
Muse Wonder acts as a heavier model that plans rules for tasks, while Muse Ember handles the volume by applying those rules across thousands of cases.
Muse Ember is the fourth model in the Muse lineup, following the Spark mid-weight model, the Glimmer open-source twin, and the Wonder frontier model.
Based on reporting by TechCrunch. Read the original article
