When I saw them say 'no more non COT models', I was minorly panicked.

While their competitors have made fantastic models, at the time I perceived ChatGPT4 was the best model for many applications. COT was often tricked by my prompts, assuming things to be true, when a non-COT model would say something like 'That isnt necessarily the case'.

I use both COT and non when I have an important problem.

Seeing them keep a non-COT model around is a good idea.