Cursor auto mode used to consume less tokens than it does in recent updates and it does the same amount of work.
This basically translates to “You should pay more regardless of your mode”, cursor is a great tool, it should be at least affordable for those paying for it.
I usually use cursor in auto mode, my pro subscription used to last for a whole month. i code daily and still will not use up my quota.
Please explain to me how my pro subscription does not last 3 days???
This is so unfair!!!
I noticed, when i open new agents, there is supposed to be a small circle measuring context usage, in recent updates, opening new agent does not mean a new environment, it passes on context usage from old agent thereby using up tokens. this will cost users unknown to then.
Please help us make cursor usable and affordable again. make the context reuse for new agents optional
Deliberately using this mode, you deliberately stir chaos into your expectations, and then you come to the forum to whine about the system’s inconsistency.
Because Auto puts some kind of model from the entire roaster under your task. It can be either Grok Code Fast or Opus 4.5… Maybe I’m thinking in the wrong direction, because I forgot that they introduced a fixed price for the token in Auto mode
Now that the Opus 4.5 is the mainstream and price has increased from Sonnet 4.5, I think Super Ultra 300 or even 400 would be in order. I could pay for that.
I am facing, the same issue, It can easily spend $10 + a day. The problem is, cheaper models are making too many mistakes, causing more issues than fixes in complex codebase, so in reality only the expensive models are useful for me. I am looking for a better solution than cursor.
Reading all the comments, I realised I am not the only one. I subscribed to $60 a month, this lasted a few days, then I added $50 that lasted a couple of days, so I just cancelled my subscription too, unless cursor makes it reasonable like VS Code pricing, I am not going to re-subscribe! Looking for other alternatives too.
You probably need to improve your prompt skills or choose a different methodology for working on the task. → Fair enough — but I didn’t need to do that until this month, and that’s exactly the issue people are pointing out here. I’ve also tried token and cost optimization, and it’s still very expensive.
I’d say my expenses since the July price change are consistent with the model I choose based on my optimization guide and the amount of time I spend working. When I was working part-time, Pro+ was enough. Now I still can’t work full-time, but I’m quite active, and I’ve almost used up my Pro+ quota in a week — which makes sense because the Ultra plan is designed for daily challenging work and I actually got quite a bit of work done.
An hour of GPT-5.2 XHigh costs around $8-10.
In the summer, an hour of Gemini 2.5 Pro cost around $6-8.
As far as I know, this is in the lower half of the middle programmer’s salary range in my country. Despite the fact that I am doing Middle+/Senior level work, albeit on individual projects.
I’ve tried GPT5.2 and Opus 4.5. They are both great models but very expensive. Grok models should be added back, waiting on 4.2 - Don’t know why 4.1 was never added.
Wow. I’ve been using cursor with sonnet 4.5 for about a week, and I’ve seen a huge uptick in just a few days! A three hour session today for a personal very small app, should not generate $120 worth of tokens!! You must be joking. The output and pricing were very different before upgrading to a larger plan. The requests, similar. I don’t know what type of perks for upgrading. Seems to be quite the opposite, twice the token load and price for very similar requests.
Again, one person, three hours, small application… over $120! I’m not building or attempting to refactor the world’s orbit!
the essence of creating a solution is being faulted here, one begins to wonder if it is just indians at the backend solving for our prompts, that should explain the exponential token cost….
ignore what the evangelists tell you, something has changed, the cursor team have just chosen to blame it on tokens, same tokens i have been using for the past year?
Same issue here also tried like adding extra $20 bucks to test and nope is so expensive, i will cancel my plan before it renews in 4 days and try windsurf (we use this at work and i use it a lot and only expended 200 requests… using only opus 4.5 thinking….) so thanks cursor you ruined it
as an update, tried windsurf it become better, and also using antigravity $20 plan you have a refresh of claude every 5 hours, and other for gemini i never reach the limits between both with cursor i used the total of the pro plan with extra 20 on 3 days…
My personal experience is that the cost has suddenly doubled or even tripled.
I have been using the Pro version for over a year, with a monthly subscription. It was working fine until November 2025, with $20 automatically deducted each month.
By December, in addition to the basic fee, an extra $100 had been spent. This price increase is indeed too steep, a fivefold increase all of a sudden. If this usage continues, it will cost over $100 every month.
I have looked at the Pro+ version, but it merely has a proportional increase in usage and price without any discounts. Only those who purchase the $200 version have discounts.
cursor must go!!!