I’m not a programmer by definition. The only actual coding I can do is some old HTML from 10-15 years ago.
But, I use and love Cursor for what it has allowed me to do. Create from scratch a social media platform (for a certain specific sport) with loads of features to bring everything together to a single platform.
I’ve worked on this project for I guess almost 2 years now. With Cursor every bit of the way (including some external ChatGPT help when models has been struggling).
Suddenly, the past week or two, I’ve hit usage limit twice. Which I’ve never done previously. Even though I’ve let Cursor plan and execute everything.
So I went from hitting the usage for the Pro subscription to being forced to upgrade to the Pro+. And it literally took 5 days until the usage limit was hit again.
I do understand that the error lies with my kind of usage. I’ve left the agent on larger models for simpler tasks etc. Maybe not creating a new chat every time I’ve needed to change something because of it being in context and I don’t have to explain everything, every time.
But still. I have huge concerns of how I am supposed to continue develop my platform if I reach limits every other day? It’s not like I can afford several hundreds of dollars every month. The Pro+ is pretty much at the very ceiling of my monthly budget. But apparently I suddenly need 10 times the usage.
All in all. I love Cursor. And I want to keep using it..
17mil tokens a day is normal. In my project, average size, maybe a bit larger, whenever I ask agent to create a plan, it consumes around 2mil tokens. I frequently run /summarize command or start new chats. If you have a chat that is near its context limit, let’s say 170K tokens consumed, a new message consumes 170K tokens just by sending the message even without agent doing anything. And when agent decides to call some tool (I assume this includes reading file etc), that consumes another 170K tokens. Each tool call is same as sending a new message into the chat (whole context reused). These tokens are charged as “cached” which is cheaper, but when agent runs 20 tools, it builds up.
I personally suggest when agent creates a plan, start a new chat, link the plan there and tell it to build it in this new chat.
Thank you for the clarification. It’s what I’ve suspected. But since the limit has never been reached for 2 years of basically doing the same thing, I assumed something has changed!
I’ll make sure to pay more attention to usage. As you said, almost at the context limit it’s kind of stupid to ask it again to change a very small mistake it did.
”The blue border was supposed to be red” etc.
Yes, sometimes I actually do give these kinds of prompts when its in context and it “understands” what I mean!
Thats its really depends on Model you use, since Cursor using price per token, its wise to use different model for different task. Example for complex, using frontier ie Opus or GPT-5.2 high but for small medium use Gemini Flash 3, or other cheapest models.
Maybe you can share ur usage? which model you use? how much context each time you use?
This help on why with 60$ you run out, even Pro+ have includes $70 of API agent usage + additional bonus usage.
