The product is not the same as i signed up for few month ago and the same shift i've experienced with Claude Code on Opus 4.6-4.7
The best way to describe the difference is you hire a reliable 'intelligent' tech lead who 'gives a shit' and in some time you eventually get over-confident junior dev that acts as destructive token burner.
No thinking during the process, not even really following instructions.
It's pretty much a binary shift that happens and in my experience can't be cured with prompting.
There is no way to actually tell what's happening under the hood with models but the difference is noticeable from the first reply.
I'm on max, GPT 5.5 extra high, always mindful of context window, use plan mode etc.
Is it the model: dumbing down or quietly route to another model for some reason? Is it the harness? or is it my imagination?
Curious what's your recent experience.
PS: frontier models should add "give-a-shit: max" along with 'thinking: max', 'effort: max' and make them actually work.