I’ve tested/used Cursor Pro for 3 weeks on a C# project. Since hitting the “fast” limit, the quality has dropped significantly:
Sloppy Code: It ignores explicit style instructions.
Duplicates: It creates new files/functions instead of editing existing ones.
Reliability: The model feels “dumber” in the long queue, not just slower.
A monthly limit that leads to weeks of poor performance doesn’t work for professional dev. I’m moving to Antigravity to test their daily quota system and agent-first approach before considering a renewal here.
Ultimately, you have to accept that using neural networks costs money. Yes, it can be quite expensive, but without it, programming is unfamiliar, slow, and boring.
It’s great that now Cursor displays usage in actual percent rather than “API cost” which didn’t tell how much usage is actually allowed.
However, looks like “Auto” is a different category now? Is it counted separately, with separate limit?
I’m sure changes in consumption and usage occured very recently, not related to august 25 because I hit limits in 2 days since last week. and I used cursor continuously.
If API costs a lot, please help us manage better context and usage
auto model these days is absolutely useless as an analytical model.
in fact across the board for ALL providers this is has been the case of a severe analytical capabilty delciine.
now the incentives are more users and reduced analytical depth to ensure more revenue.
Compared with the peak model, which was probably the first version of Gemini 2.5 Pro. Current day AI models are a joke. Yes they have a lot of breadth and context, but they have NO depth any longer so that means that inference and complex problem solving (real world! not benchmarks) has been severely diminished.
I have noticed this first hand with my large mono repo, it’s now near impossible for any AI model to do complex analysis without just getting lost in the parameters.
Unfortunately, to work with AI these days you have to do a lot of the grunt work of problem analysis, work separation before you have the AI do anything. That inference power is gone
I too think its sad that unlimited Auto has disappeared. However i understand the proposition. A lot of users (i would expect 10 thousands of users) are abusing the auto mode. This is not a sustainable business model. I have seen pictures of 1B+ in token usage. this essentially means they are running their models constantly free of cost, making that magical tech dept. And i believe there should be ToS’s and rate limiting instead of this approach to removing the Auto mode. Allow users to use it, at a realistic amount. The limit now is probably very profitable at the usage rates people are reaching. But as some people mention in here, at least give options to people who don’t abuse the free usage to continue to use cursor at more generous limits. I think the goal should be to be to leverage the larger customer base of cursor. instead of just ruling out everyone who use it. That is essentially giving away all the users to Antigravity that gives more generous usage limits.
Im staying with cursor a while longer, even upgrading to 3x limits for a while. But i hope there comes some benefits to being loyal to cursor over switching to antigravity.
