Cursor Performance Pro vs Pro+

Ideas Feedback
1

I started on Cursor Pro and switched to Cursor Pro+.

The quality of what came out since I switched is dramatically worse.

To a point where I’m considering not using Cursor anymore, try something different because this just result in very frustrating experience of fixing one thing to break another and once fixing the other going back to the original thing in an endless loop. I can see the tokens being used with very little to no value from my point of view since I have to repeat the same process several time a week.
No matter what I do we keep getting into the same loop, what should be simple and addressed once seem to be just a temporary fix and come back a few days later. out of frustration I’m now saying my prompt in NotePad++ so I can just copy and paste: that a good indication that the root cause is not addressed. What I see is lipstick on a pig. Just enough to say “done” for the day and burn the tokens for another round tomorrow. I questioned several times the validity of the end result, ask at several iteration to look deeper, adding self-testing loop with the desirable outcome.

The pattern is coming somehow predictable, when asking to make deeper research I know it will come back in a day or two to hunt me: at some points why do I pay the extra $40 when the tokens just burn up in the front of me.

I’m only using 34% of my usage, but most of that is the frustrations to see the same issue coming back later. Those are all home projects, nothing that would break a production server: yet it’s very disappointing. We are not speaking of anything incredibly complicated either.

I didn’t see such pattern prior to the Pro+, maybe it’s a coincidence but I doubt it.

6

Hey @Deckard,
Quick reassurance: Pro → Pro+ doesn’t change which models you get or their quality - it only raises your included usage. The extra $40 buys more usage, not a different model, so the upgrade itself isn’t causing this.

The fix/break loop is usually workflow-driven. What helps most:

  • Start a fresh chat per task - long sessions pile up context and make the agent go in circles.
  • Pin one model (vs Auto) so you can compare like-for-like - available models.
  • Add a rules file and ask the agent to self-test before calling something done - that’s usually what breaks the “temporary fix comes back” cycle.

If you grab a request ID from a bad response (three-dot menu → Copy Request ID) and tell me the model, I’ll look into that specific request: how to find it.