First impression of gpt-5.5 is that it's just as bad as 5.4 at not taking initiative.
You have to make sure to not phrase your prompt as a question as it will just answer it instead of fixing what you are questioning.
So in summary.
Is GLM-5.1 an Opus level model - no
Would I trust it to write its own specs - no
Can it generate good code if you understand its limitations - yes
In the end I split the process down so now I have
Write story - GPT-5.4 high reasoning
Implement - GLM-5.1
Verify - Opus 4.6
This setup is now able to work for much longer before falling over. It's still GLM which trips up but it's less frequent.
@VicVijayakumar As others have pointed out - in the UK for every £2 you earn above £100k you lose £1 of your nil rate band.
You don't decline the pay rise though. You just increase your pension contributions to take your taxable pay back below £100k.
Same take home pay, more retirement funds
@hamen This is why I switched from Cursor to Claude Code. Even on the £90/month plan the vibing never stops for me. Nonstop Opus.
You do lose the tight IDE integration but I've been using inside the terminal inside AS/RubyMine/Pycharm as required