deepseek-v4.1IMAGE
DeepSeek V4.1 Flash scores 42.7 on a 105-bug repair test
On 105 real bugs from two overseas projects, Kimi K3.1 Pro scored 21, DeepSeek V4.1 Flash 42.7, and Opus higher, undercutting the idea that pricier models perform better.
洋仔实验室
@yang3izcArchived 4 days agoOriginally posted 2026.10.26
洋仔实验室 @yang3izcOriginally posted 2026.10.26
贵的不一定好用。Kimi K3.1 Pro 修 bug 只拿到 21 分，DeepSeek V4.1 Flash 42.7，Opus 更高。105 个真实 bug，两个海外项目实测。我原以为贵有贵的道理，结果榜单直接打脸。测法来源在哪，有人扒到原始仓库吗？
x.com/yang3izc/status/2114477211717927313Archived 2026.10.26 12:05