That’s is exactly what happened, they specifically raised the fees they charge Cursor.
My view is different. We are a small development team we spent fortune on contractors helping with coding.not mentioning the trust issues. Now I can do our developemnet alone 90% of the time. We are saving approx 1000-1500 a month. I’m more productive with with Cursor and our development timeline sped up even though I’m doing it alone. So I’m more then happy to pay around 100bucks a month. ( saying that the most I payed was 30 at this stage).
So, yes it’s getting more expensive but I’m still way in the positive mark and I’m quicker.
@ROBERTO_CHAMBULE Do you write your own code, or does the cursor do all the work? If you use Claude, then yes, you’ll run out of credit very quickly. (Honestly, I use Kimi v2, it’s really cheap and works very well, it understands what you ask it to do.)
With Kimi v2, I make 20 times more requests, and I can assure you that it understands exactly what I am asking it to do.
@danperks just one doubt: I noticed the image you shared shows a bonus of $5, while I got a bonus of $30 last month. Do people get different amounts of this bonus value, and if yes, what does it vary based upon?
It’s being consumed too quickly. Last month, it reached $51 before giving a warning, and now at the beginning of the month, it’s already close to the $16 limitIt’s being consumed too quickly. Last month, it reached $51 before giving a warning, and now at the beginning of the month, it’s already close to the $16 limit
Cursor is kinda scamming with usage pricing cuz companies, obviously, get better deals for cheaper apis because they buy in bulk but yet they want users to use usage based pricing which is just so much more expensive than what it costs for them.
Hi everyone. Here are more details on usage:
- We show you what is reported by AI providers based on your consumption.
- High token usage means higher consumption. Also heavier models cost more.
- One request can be 100 tokens or 1 million. which can be < $0.01 or > $100.
- Make sure to use focused and shorter chats as over many tool calls and follow up requests the token usage accumulates.
- We are improving Auto models, they will get better as well.
- Thinking models usually do not perform so much better than non-thinking models but they consume more tokens.
- Avoid attaching files, rules, MCPs when not necessary as they add up context usage.
More on token usage, how it affects your consumption and how to optimize it:
The auto models are definitely better than they were even a week ago, I like the changes you guys have done. The auto models instantly beat gpt 5 without thinking, even hours of gpt 5 without thinking got beaten in 1 auto request.
Except Anthropic jacked Cursor’s price which is where all this has come from
I was trying to read the article to learn more about the industry but I cannot find a valid argument or proof. I was even going back to his references, as he deemed them necessary to understand the blog post. Nothing. No argument made, no proof delivered – at least from what I can see. I was also checking other links, but they are often just random statements from random other blog posts without proof.
Can you please help me out and quote what you are referring to? Where can I read how and when Anthropic cranked up the pricing and is bleeding out its enterprise customers, like cursor?
I like the Claude models.
If I ask Cursor to change a UI element, it takes one request.
If i ask it to change 5 UI elements (in one prompt), it takes one request.
Why would I want to use multiple short requests?
There seems to be some misunderstanding. You can combine changes that are related and ask AI to change 5 elements in one prompt.
However Cursor is not tracking requests but token and model usage instead.
I mentioned using shorter chats. That means when you finish one task and start next task that does not benefit from all the previous context using a fresh chat avoids reprocessing the chat context from previous task.
My cursor expired on my PRO plan a few days ago. I didn’t renew. I liked the plan I signed up for at the time.
500 requests + slow requests was a perfect balance.
Now, with PRO+, I can’t use it and have the same productivity.
See you soon. If cursor prices improve, maybe I’ll come back.
I usually last like 3/4 a month on my quota. But remember: It’s not over when your quota is reached! You can still use it forever in auto mode!
You would be surprised at how much it can still do as long as you design your specs, unit tests, and task well! In fact, I always launch my Background Agents in auto tier (you can choose it if you do from mobile), and it does a good job filling in the rest of my unimplemented specs.
Try also adding in cursor rules to tell it to be concise with it’s response, in hopes of reducing token and context usage. Most answers can be answered in about 2-3 sentences, instead of paragraph posts.
Guys and gals, do you understand that Cursor resells LLM models? If a model costs $10 per 1M tokens, Cursor can’t charge for it $5. You can complain as much as you can, but it’s just business. If you don’t like prices you can write code on your own. I also don’t like to pay, but it’s just business - nothing more. The fact it was cheaper before - Cursor had negative profit to let you try and feel what it is coding with LLM. So, you decide use it or not. Sorry for the truth.
Honestly I never run out of auto requests. Compared to Claude Code I think it’s a bargain
What I do is generated major code using claude code and check errors and smaller features using cursor auto mode. It’s perfectly balancing for me. Yes cursor and claude also need to make money. Obviously they want profit too. If u want to use, use it with their pricing. don’t simply hope that they will think of coders profit. They have to answer their investors.
Ditto, hit the same problem.
While I understand the way cursor team tried to explain, it can not be a after-tokens-are-out discovery for us as it is disrupting the monthly planning to build and release the products.
Yup agreed they lost me as a customer now as its too expensive for how good it is.
Shouldn’t that be taken care of by “auto” ? This seems like proof of price gouging - We have models that can answer your question cheaper and faster, but Cursor decided to charge you more and do it slower.