Because those are the cases that still break, and demos are marketing rather than evaluation. If you want the honest picture you have to generate the hard prompts yourself, which most people never do before forming an opinion.

text on a sign is the real benchmark and everyone knows it

EXACTLY 🔥 put it in the demo reel cowards

哈哈这个说得对，招牌上的文字基本是照妖镜，一测一个准