Meta introduces Muse Ember for low-cost AI agent tasks

Meta has released Muse Ember, a lightweight multimodal AI model aimed at the repetitive work performed by AI agents, including tool calls, document extraction, routing, coding tasks, and long-file analysis.

The company positions Ember as a lower-cost model for high-volume workflows rather than as a replacement for its most capable systems like Muse Wonder. Meta says Muse Wonder should handle complex planning, while Ember carries out large numbers of smaller tasks. TechCrunch likewise describes the model as a component for agent systems in which a larger model sets rules and a cheaper model applies them across many cases.

Muse Ember costs $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 200,000 tokens. Meta also offers lower pricing for cached inputs. The model supports a context window of up to 1 million tokens and can generate up to 512,000 output tokens.

Benchmark claims and capabilities

Meta says Ember scores 36 on Artificial Analysis’s Intelligence Index at an estimated $0.14 per task. The company claims this places the model ahead of competing low-cost offerings on the benchmark’s cost-to-performance comparison. These results include comparisons with Anthropic’s Claude Haiku 5.5 and OpenAI’s GPT-6 Luna, though benchmark outcomes vary by test.

On coding evaluations, Meta reports that Ember beats GPT-6 Luna on Terminal-Bench 4.0 and Coding Agent Index-AA, while Claude Haiku 5.5 leads Ember on FrontierCode 2.0-AA. Meta also says Ember can process text, images, audio, and video, allowing agents to work with screenshots, scanned documents, recordings, and video.

The model is available through the Meta Model API and Muse Code under the name muse-ember-1.0. Meta says it evaluated the model for chemical, biological, cybersecurity, and loss-of-control risks, and found it remained below its internal high-risk thresholds.

Sources

• Introducing Our Smallest, Most Efficient Model: Muse Ember – Meta AI
• Meta releases Muse Ember, a new lightweight model for agent execution – TechCrunch
• Meta’s best AI model now trains the one most of its users get – The New Stack