Just use Cursor IDE with Sonnet 4.5 and you’ll get more almost identical .md files than you get code. Even when you have a rule that tells it not to create them.
Steps to Reproduce
As mentioned above. It’s the last 2-3 weeks this is crazy.
Expected Behavior
Not to generate insane amounts of similar .md files.
I am honestly starting to get a bit annoyed with the poor support. Used to be good, now you are just chatting with the AI on Email and when you’re not, then you speak to someone that just talk about something completely different and don’t get the situation - even though forum request you to contact support ref. refunds when a bug causes extra usage…
I spent about 2k usd last 10-14 days … many other users just create new Ultra Packages when they’re used up to save funds… You want people like me to do the same? I start to consider it when your support is so poor… Why would I contact support for fun?
It’s like we as customers pay extra for your errors/bugs and then we help you guys sorting them and WE have to pay for it… Why the … would we do that? Your support people live on a different planet or something …
2k lines of “code” on .md files and 200 lines of actual code…
Support: “It’s not a bug, it’s very normal. It is what you requested” … Yes it surely was I like to wait 10x longer for something I just delete afterwards…
Note: Not talking about support on Forum which is always quick and nice. But that email support.. After last message I just got ghosted, such a joke.
Okay. So what models can we actually use currently? Opus 4.5 is also bugged, in a similar way, plus it keeps sumarizing chats forever, and ChatGPT 5.2 High Extra is also going in circles and just spending forever until you manually stop it.
So is this a Cursor problem or ? ….
I am sorry, but I am not really interested in helping anymore cause the result is just time wasted and often money on my behalf.
Next I am gonna test is Windsurf. It’s annoying now guys, sadly.
I get the frustration. When multiple top models behave unpredictably, that’s a real problem.
On models:
GPT-5.1 High: the community says it’s the most stable right now. It’s also cheaper than Sonnet and Opus, and it’s reliable.
GPT-5.1 Codex Max XHigh: also looks pretty stable.
GPT-5 Mini / Grok Code Fast: good for simpler tasks.
About GPT-5.2: there are reports of higher token usage, and the team is looking into it. If you have a Request ID from a GPT-5.2 or Opus 4.5 loop, please share it. That helps us prioritize the fix.
I’ll pass your feedback to the team about email support quality.
I get it if you want to try alternatives. Nobody should be paying for bugs.
Thanks for getting back. Another thing I’ve been running into many many times now as well is for 4.5 Sonnet you say it has 1M as token limit. That is simply not true. It stops ALWAYS around 400k and just before…. And then it stops in the middle of tasks and then you cannot open a new chat and then refer to the other that just messed up. This is really bad ….
Yea I recall 5.1 High works question well. Might have to switch back to that.
But you hear how stupid it sounds that people report high token usage and bugs etc. and our wallets get drained, normally a business would refund when it’s their fault. This is mad to have your customers paying for a product, then it bugs and the customer pay for that as well - along with trying to help to solve the bugs… Sounds like some free comminuty project you just charge obscene money for …
4-6 months ago I was charged waay too much as well and back then I spoke to a helpful guy, but the last interaction with support is just the most horrible… I have after all spent around 10k usd with you guys now.
You recommend to use planning mode? Do you know planning mode doesn’t work either currently …. It takes 1-2 sections for the model to write 1 word … then planning mode timesout and funds wasted again - AS USUAL!
Yet I am here paying for bugs and errors, because Ivy doesn’t bother to answer or look into the case properly. When I am told to contact support and then reach a person that don’t bother to check what the case is about.
I have stayed away from mentioning the actual name, but now I got enough… So your person Ivy on support is just sending useless and off-topic answers and avoiding the problem…. Very annoying to be honest.
Honestly… what is going on here the last 2 weeks time … now plan mode do not work. It Doesn’t finish tasks, it just creates a random .md file when asking it to build and then follows that halfly-randomly. Then you ask it wtf it’s doing and it’s like ohhh sorry I did not follow the plan I just made another and implemented some of that…. Have you messed my account up because I pinpoint that you guys are going in a very wrong direction???
If this stuff continue and with your support not getting back it’s gonna end with chargeback on what I have been unreasonably charged for. Not much, but some for sure….
I agree with everything - I usually don’t exceed the monthly Pro subscription limit. However, for December, I received numerous .md documentation files and have already spent over $50 in addition to the subscription. I didn’t have this problem before.
Something is going on for sure … The main problem is they just come up with excuse and blame you as a user. They did not use to do that and then they ghost you on support… 12 days now, nothing… I don’t know but I would say 300-500usd here of usage due to errors etc… out of 3500 usd used + 200(400) on ultra subscription.
