Osegi
AI·1 min read

Meta Muse Ember: Small AI Model Aims to Make Agents Cheaper

Meta's smallest AI model, Muse Ember, matches Claude Haiku 5.5 on the Intelligence Index while costing only about a third per benchmark task.

Abstract illustration of a compact AI model chip glowing in Meta's blue colors next to a falling cost curve.
Image: heise online
SUMODialoediviewergiftdetect

Meta has introduced Muse Ember, its smallest AI model so far, and the company is pitching it as a way to make AI agents considerably cheaper to run. On the Intelligence Index, the model reaches the level of Claude Haiku 5.5.

The cost difference is the notable part: per benchmark task, Muse Ember costs only about a third of what the comparison model does. That puts a small model on par with a well-known rival in measured performance while sharply lowering the price of each completed task.

Why it matters for agents

AI agents typically work through many steps and model calls to finish a single job, so per-task costs add up quickly. A model that delivers comparable results at roughly one third of the cost can make such workflows more affordable, especially for applications that run at scale.

The figures come from a benchmark comparison, so real-world savings will depend on the workload. Further details on the model, such as availability and pricing, were not immediately available.

Apps worth a look

SUMO: the number puzzle

Tap the right tiles, add them up, hit the target. Easy to learn, hard to master. Free, for all ages.

Tags:metamuse emberai modelsai agentsclaude haiku

More from this day

See all →

More in: AI

See all →