Recently, after cursor was upgraded to 2.0, the token consumption has decreased by 2 to 3 times compared to before.
I thought this was a good thing, but then I found that the model’s intelligence has dropped significantly. Problems that could be solved in one or two conversations before now require more corrections.
If intelligence is used as a trade-off for the reduction in token consumption, I’d rather you go back! Because this has wasted more of my time and energy!
Model intelligence dropped to a level that this becomes mostly useless. If previously it was a good educated junior developer, right now it’s a mid-school student forgetting (and sometimes simply ignoring) instructions, going with drastic uncalled changes, completely ignoring best practices, introducing really bad code smells left and right, etc.
I disagree, for myself personally and using the models I am using with android, the reverse has happened and everything seems to be much more efficient. I think it all depends on what you’re actually trying to achieve, you can’t base this off one model.
It’s not really about the model itself. Yes, it clearly falls back to the cheaper models now, but that’d be manageable. The issue is with the cursor internals. It now effectively ignores most of explicit instructions via .cursorrules, ignores instructions directly from the chat when summarizing context, etc. As the original poster mentioned - it seems they’re over-cutting token consumption leading to this bringing more headache than value.
If previously I was always falling back to cursor after trying other tools, because of extreme value (for me at least) of the wya they implemented changes review - now I find myself almost not using it in favor of other tools.
I have not been loving Cursor’s new personality. It’s a lot less conversational now and less friendly. I understand this is meant to be a coding tool and not an AI girlfriend or something, but the previous “more human” version actually led to much better decisions and thinking. Now it’s a lot dumber. Can’t make good decisions. Can’t figure things out. Just blindly charges through whatever prompt without any care for bugs or best practices. I wish I could roll back to Cursor 1.0.
