If you’ve maxed your API monthly spend, and you switch to Auto - which is unlimited - which model will it use in this mode?
You don’t know which model is selected - but basically you can directly ask “what model are you?” in chat.
If we can see which model is selected in IDE and why the model selected explaining in IDE will be still reliable in some cases.
auto = gpt 4.1
4 Likes
It’s not always gpt 4.1
Auto-select
Auto keeps you in flow by picking a reliable model from the pool above (excluding
o3). It does not route based on task type, but is a solid default if you are unsure which to choose.
AI models get cheaper and faster over time, not Cursor apparently it gets more expensive by the year even though sonnet’s api costs are the same.
Because this company is more prone to committing fraud than trying to make ethical choices in their business. Be wary about any data you send them.
2 Likes
“Abuse the system” … you realize auto changes models and messes up context constantly right? Also anthropic is $20 for unlimited claude 4 with clear rate limit timings. Auto mode is a joke.
2 Likes
So, can we still trust the privacy policy they claim?
Thank you for the clarification on “local limits.”
It’s helpful, but the core issue of predictability remains. When I need to plan my work, the question “Can I finish this feature today using the local limits?” still has no clear answer due to its “best-effort” nature.
And perhaps, that’s the point. I’m starting to think it might be unreasonable to expect a $20 Pro plan to reliably power a project that generates hundreds of dollars in usage. For that level of certainty, we probably should be paying more.
This leads me to speculate that the “local limits” after the $20 credit are intended for minor, non-critical tasks—like generating git commit messages—rather than core development. If so, their unpredictability makes more sense, though it feels a bit like a “chicken rib”: a bit underwhelming given the powerful models available.
Personally, I remain optimistic. The Pro plan has provided incredible value this month, saving me hundreds of dollars. If this “black box” continues to be this generous, I can personally accept it.
However, to satisfy the needs of the entire community, I still strongly hope we can move towards quantifying and visualizing these “black box” issues.
Keep up the great work. I’m still rooting for you!
2 Likes
are you sure?
One would think that this feature should have been shipped much MUCH earlier.
I have no clue at present as to when I’ll hit my limit. So to hell with planning to do anything with cursor, they’ll cut you off seemingly at random and force you onto a garbage model instead.
1 Like
The Cursor team’s wording really annoys me.
At this point, just create progress bar that represents the remaining credits, when recharged, displayed on that bar too.
I understand the cost of AI is increasing but let’s be transparent here please.
9 Likes
With the new pricing system, just saying “Hi” to Cursor costs me $0.22.
Really, guys? Do you honestly think this is going to work?
12 Likes
I tried the recommended auto mode, it’s so bad! Ruined my projects, I had to restore from backup and vow never to use it again. Sonnet got used within 1 day and now it always just says “you have hit the rate limit for this model”, switch to “Auto”, which I am never using again, in a matter of 3 prompts it destroyed multiple parts of my project. Before at least we had unlimited slow requests. I think I am done with Cursor, there is no other choice, I will have to learn claude code.
4 Likes
Simplify, simplify, simplify - its kind of obvious if you want to survive this.
[Edit - T1000 - translated text to English]
Greetings, could anyone kindly explain the concept of Cursor?
Been using cursor for months without that many issues (other than when they had the slow pool issues) and I’ve always been able to do my month of work with it.
My billing cycle started yesterday, 5 hours of work in with sonnet and I’ve used more than 60% of the monthly included usage, that’s rough.
For a brief period we were able to see the price of each request in the dashboard instead of “Included” that was great, why has it been changed again?
Edit: Tried Auto and god I ALWAYS need to tell it to do it’s thing. If I don’t say “Make a todo list and do all the steps” it won’t even use the todo list and will do one small step and stop there.
Something is WRONG with how Claude Sonnet 4.0 is using TOKENS!
Same project, same tasks!
5 prompts and I hit the rate limit
Switched to gemnini 2.5 pro which is acceptable for now, way less TOKEN usage!
Two issues I have at the moment:
- I cannot see my actual current usage, meaning: how many credits do I have left?
- The docs talk about a pro+ plan for 60/mo but the pricing page does not show this plan:
