4.6 has been fantastic for me
I've tried the latest models across OpenAI, Claude, Chinese, etc. They just do stuff. That's not how work is though. You want them to do specific work, at which point it's a real hassle to follow up on all the garbage they have been outputting.
/long rant