模型50

Muse Ember 实测纠错需求更低

原标题:Based on thousands of real-world user sessions, Muse Ember’s key strength is results that need little to no fixing.

Arena.ai@arena发布时间 (北京时间)

基于数千次真实用户会话,Muse Ember 的核心优势在于结果几乎无需修正。 引导分数(越高 = 用户纠正越少): - 所有有经济价值的工作:+11.2%(Luna +3.9%,Haiku +3.8%) - 软件与 IT 服务:+11.6%(Luna +5.0%,Haiku +6.3%) - 涉及文档的任务:+12.0%(Luna +0.4%,Haiku -0.1%) 差距在文档类工作上最大,Ember 在此优于同类产品。

中文译文尚未完成,先为你展示原文。

Based on thousands of real-world user sessions, Muse Ember’s key strength is results that need little to no fixing.

Steering score (higher = fewer user corrections):

  • All economically valuable work: +11.2% (Luna +3.9%, Haiku +3.8%)
  • Software & IT services: +11.6% (Luna +5.0%, Haiku +6.3%)
  • Tasks with documents: +12.0% (Luna +0.4%, Haiku -0.1%)

The gap is widest on document work, where Ember outperforms peers.

Agent 智能体评测基准#评测/基准#Agent