浏览器与 computer use 任务：Luna 96.5% vs Jev 37.3%

实测对比显示 Luna 在浏览器工作流和 computer use 任务上成功率 96.5%，作者找到的最佳 Jev 实现只有 37.3%。

Steve (Builder.io)

@Steve87082026.09.24收录原推发布于 2026.09.23

浏览器与 computer use 任务：Luna 96.5% vs Jev 37.3%

Steve (Builder.io) @Steve8708原推发布于 2026.09.23

I ran a set of basic browser workflows, more real-world end to end tasks, and some computer use tasks Luna was able to accomplish 96.5% overall, while the best Jev implementation we found succeeded only 37.3% of the time

x.com/steve8708/status/2102534746110906661采集于 2026.09.24 10:19

在 X 上打开原推