GPT-5.5 did a significantly worse job than Qwen-3.7-Max on a job today (some devops tasks I wanted to create some reusable scripts for). Kind of disappointing.