WHAT NEW IN X11.06 FRI

AI collections from X · 23 collections · 69,408 posts

jevIMAGE

Claude Code vs Pi + Qwen 3.8 tested on a real acceptance pipeline

The author ran a Computer-Use Agent comparison on a real acceptance workload using Claude Code with Sonnet against Pi with Qwen 3.8 27B at default thinking depth, and suggests DiffusionGemma as a Jev executor as a direction for improvement.

WeZZard
@realWeZZardArchived 2026.09.27Originally posted 2026.09.26
Claude Code vs Pi + Qwen 3.8 tested on a real acceptance pipeline
WeZZard @realWeZZardOriginally posted 2026.09.26
Open original on X

这是我自己验收管线上的一个真实工作负载,使用 Claude Code 加 Sonnet 作为 Computer-Use Agent,对比 Pi 加 Qwen 3.8 27B 默认思考深度的结果。 当然,Qwen 3.8 27B 还有很多可优化的空间,比如说降低思考深度、累积数据集后进行 fine-tune、引入 DiffusionGemma as Jev 作为 executor。 感觉这个模型已经到了可用的阶段了。

x.com/realwezzard/status/2103745507507999198Archived 2026.09.27 10:12