Hi Folks, is there any update? Would be great if we can get some communication from the Cursor team. 
Any news about sonnet 3.7? I tried with api too but doesn’t work. I tried with others and it work. So it depends from Cursor? Thx team for all. I love Cursor 
hi,
howcome you put slow pool users the SAME as free trial?
why i paid the 20USD/month then?
i’ll suggest you triage into
high speed pool
slow speed pool
free trial pool
indeep.
otherwise i’ll take my own action - i quit.
thank you
5 Likes
some times it’s sad that there are so many poles,
that facebook’s olllama and google’s gemini are full bottles awaiting for nothing.
I’ve noticed over-consumption of my fast requests when using the Resume this conversation button on 0.46.7
It appears that EACH TOOL CALL uses 1 fast request after clicking the Resume button. This made some of my prompts going up to 26 fast requests being used (1 for initial prompt, 1 per tool, limited to 25), which is unconceivable.
This happens at least with Claude Sonnet 3.7 Thinking, in Agent mode
Be careful with that - hoping Cursor team reacts quickly to that. I think that a lot of people may have experienced the issue and should be refunded.
Please, this is a critical issue for the ones who didn’t notice.
11 Likes
In Win… free for free users… and a free trial, what is this limit in Cursor? This is not professional, and what is happening is not acceptable. In Trae also free. Too bad, Cursor, you disappointed me
2 Likes
This makes no sense. Cursor should re-negotiate with Anthropic. If the business relies on a certain availability of servers then the supplier can not just start bricking all end consumers if they release some new models and features.
complicated because my project is very large, claude 3.5 can’t edit or fix any bugs or add more code because it gets completely lost in the code and starts deleting a bunch of stuff. it literally starts hallucinating. with 3.7, i’m still able to edit, fix bugs, and do a lot of things in my program… but unfortunately, it’s offline ='(
any forecast for when 3.7 will be back and working properly?
1 Like
Is anyone else finding that both Claude 3.7 and ChatGPT 4.5 seem worse when chatting with codebase? I’ve been trying to for an hour to write and debug some tests and they seem to be really good at avoiding the underlying issue.
I have found it to be an issue with 3.5 too, often diverging from the root cause. Interestingly, Claude Code seems to be coping better with those cases.
Havent tried Claude code yet, are they going in the right direction?
Claude 3.7 is a mess for me. It enters infinite loops (apply and reapply to the same file repeatedly).
So many times it hits 25 operations cap, and it could not fix a minor issue in the code for 20 minutes. Switch to o3-mini, find the problems, and fix in one step.
Burn almost 200 requests and solve nothing.
BIG, BIG disappointing for me.
2 Likes
the big problem may be that you ask very broad questions, so the ai thinks and analyzes much more.
depending on the size of your program or file, yes, it will go well beyond 25 operations.
personally, i managed to reach around 6k lines of code in a single file, and after that, version 3.7 started to hallucinate a lot. after about 10 to 20 attempts to find a simple solution sometimes (i was testing).
for small files up to 2k lines of code, it’s wonderful… if it goes beyond 3k ~ 4k, version 3.5 already hallucinates. now, version 3.7 still finds solutions. after 4k~8k, only 3.7 might be able to help you find a solution or implement something for you.
There’s nothing that can be done. They stated it’s a limit placed on them by Anthropic. You need to consider alternatives if using 3.7 is important to you.
until it resolves maybe you can try to use gpt 4.5 on cursor.
Bedrock is a solution
hi,
PAID users in the slow pool also? please state clearly thx.
I am a pro user who used up the 500 fast prompt already.
1 Like
Why is Claude 3.7 listed in Cursor – Models while GPT 4.5 isn’t?
Just wait for the Deepseek 2 team to stop distiling it and we can have access :'D
I have paid subs to Claude and OpenAI pro … can I configure cursor to use my accounts instead of my Cursor account for completions within Cursor?