用户反馈：GLM 5.3 比 V4.1 Flash 1017 更慢但更聪明

Yume_X @yume_arasaki原推发布于 2026.10.24

在 X 上打开原推

Benchmaxxing is a really big problem....does anyone have any idea of how to measure models better? I have a feeling same deal with Orninth, it's probably not much better, or maybe weaker than Qwen 3.8 Flash Moonshot has always been like this. Also i'm finding GLM 5.3 to be slower and smarter, than Deepseek V4.1 Flash 1017 https://x.com/yume_arasaki/status/2113782528473706566?s=20

这条是引用着另一条写的：x.com/yume_arasaki/status/2113782528473706566