i’m trialing this, and getting alot of
“We’ve hit a rate limit with Anthropic. Please try again in a few moments.”
so - should i pay for this? or just pay directly to claude to avoid hitting a probably shared rate limit?
if so - i don’t understand whats the purpose of cursor 20$ monthly if the quota is shared
No, there’s currently a bug where even if you pay, you only get slow requests.
No for - not worth buying?
or a bug that should be resolved and my prompts won’t hit limit?
also - 500 prompts are months seems low
To give you a better sense of the issues with your question :
The paid subscription allows for a not easy number to track of coding sessions with magical fast code assist. Unless it is not magical.
When paying, Tab to complete feels like the only editor you want to use, or I would dream of Apple Intelligence giving me this in all apps with my private data sort of magic. So if you code long hours, you will enjoy and get used and it’s awesome.
When you hit your limit, you still have the slow requests and smaller models, the experience looses the magical touch, your mood, productivity is severely impacted and In solid 3 months experience (10+hours daily) you will either accept it with a sad face and complain here and social media to relieve the pain, or just realize it’s worth it to pay more to keep your accelerated and fun productivity.
but here is the catch, Unless it’s not Magical
more and more, I see a variable quality of output from the same LLM dealing with the same problems. To be clear, this is what I expect from humans, after hangover, vacations, accident. And took me a while to learn to be alert on the output, always, because it seems to vary wildly and to summarize: there are many good days and a few terrible days.
And I experience this with at least 3 different paid LLM subscriptions ( ChatGPT Plus, Google Gemini Advanced and including Cursor Pro here, even if it is different) .
I am not sure of the cause, I have a really consistent prompting ( that improved ) and love to structure a good question, and I can only suspect the Platforms Inferencing cost, like all cloud services have their SLA/SLO, Capacity that needs to be managed, so when you have a bad days, likely there is less memory/gpu/time/etc which reduces the ouptut quality.
specially with code, as Cursor mentions somewhere in the docs, limiting to 20K tokens most of the time ( if LLM supports it ), for code we need a large window ( larger than the input box ) and this is of course one of the aspects that can be optimized.
If you are aware of those, accept them you will be a happy customer, until something better appear, because while it’s good, it’s rather unconfortable to be vigiant and loose entire day of work when trying to solve a bug, creating new dozen and in the next day same the llm prompt problem, solves in 2 turns.
hope this helps,
Páris
and after I clicked retry, new turn, new code suggestion and a good example of a bad day :
focused change, was a major destruction of the code logic in the file I am working a py file with less than 200 lines, being a third of them logging debug and print lines, another third comments.
and I am using what was considered the best LLM for this task Claude-Sonnet
following up :
and I hope it’s clear of what a bad day is like.
BTW , the change is targeted at lines 64-84 , and I would usually click apply and run, but not anymore, because this sort of problem when applying changes is the worst and random.
But I welcome any tips to make this easier, maybe I am still learning.
I am getting this too, It seems something is wrong! 390/500 is shown on my panel. I no longer would recommend this app to others.
Hey @kevin6! I’d encourage you to reach out to hi@cursor.com so we can take a closer look at your specific account!
I hear you! For precise edits like this, I like to highlight the exact code and hit CMD+K - keeps things focused and tends to work way better than trying to explain what you want to change in chat.
There are now so many posts on this. The Cursor team keeps saying ‘email us’, but they don’t actually respond or do anything with it.
i Think i’ll give it a go for a month, i do have a professional chat GPT account and i just cancelled claude, if this doesn’t improve i’ll just switch back to calude and dump cursor
We sure are reading it and responding as fast as we can@hococib347! It’s just that we have a lot in the inbox to go through
