AI TrackerGet the app
← All news
MetaModelsCoding

Arena: Meta's Muse Ember leads efficient models in Agent Arena at 1.9¢ per task

Arena says Meta's Muse Ember (Max) is the new leader among efficient models built for real work, ahead of OpenAI's GPT-6 Luna (Max) and Anthropic's Claude Haiku 5.5 (High) on every measure in its Agent Arena:

| Model | Net improvement | Cost per task | Time per task | | --- | --- | --- | --- | | Muse Ember (Max) | +2.3% | 1.9¢ | 27 s | | GPT-6 Luna (Max) | +1.4% | 7.0¢ | 5.8 min | | Claude Haiku 5.5 (High) | +0.4% | 6.9¢ | 1.9 min |

Arena calls Ember about 13x faster than Luna and less than a third of the cost of either. It also points to an analysis of Ember's steerability score on work tasks and to the real-world tasks users are running with it. These are Arena's own figures; an earlier Arena report had Ember at #25 with a +1.7% net improvement.

Get news like this as a push.

AI Tracker alerts you when something changes for the tools you use.

Get the app