I’ve been using Auto for months, and something unusual has started happening this past week. Normally, after a full day of using Cursor, my usage percentage would barely move, maybe one percent at most.
Lately though, with the same workflow and the same Auto usage, my allowance is climbing much faster. It is burning through a noticeable chunk every day, which has never happened before.
I have not seen any announcements about model cost changes or Auto adjustments that would explain this. The only usage update I remember was back in August, and those changes kicked in at the next billing cycle. My billing renews monthly, and it is already December, so I am confused why things would suddenly spike now.
So I am wondering:
Did I miss a recent update about Auto or model pricing?
Is there a setting or behavior that could make Auto consume higher credits now?
Has anyone else noticed the same thing?
I would really appreciate any transparency or clarification so I can understand what is going on.
Auto mode is priced at 1.25 for cache write 0.25 for cache read and 6 for output, you can see the other model prices here:
for comparison:
GPT-5 is 1.25 input .125 cache read and 10 dollars output.
Grok Code is 0.2 Input 0.02 cache read and 1.5 output
I don’t know of any change in auto pricing recently, is it possible you are linking much more context data than before? this will cost you more. The only change I know of was that auto was free a while ago but they transitioned to this pricing model a few months ago.
Dunno what you’re talking about, this is totally normal and not at all out of the ordinary for the output which A____pic’s tool produced in 30k tokens for < 1000 LOC and no tool calls. I can’t actually tell if the billing or their configuration is broken, but I don’t think they can either.
Genuinely, it’s impossible to tell if the billing is broken, there is creative accounting, or for some reason they thought no one notice that what they’re appending to prompts is causing 10-30x previous usage.
This is very troubling stuff as I’d hate to see Cursor slowed down by everyone going to their bank or credit union and processing a chargeback due to billing and usage irregularities not being experienced anywhere else.
You’re definitely not alone in noticing this – many users have reported the same sudden increase in credit consumption over the past weeks, and it’s not your imagination.
What’s happening:
“Auto” mode is no longer unlimited as of September 15, 2025 (or your next billing cycle after that date). Since you’re on a monthly subscription and we’re now in December, this change has fully kicked in for you.cursor+1
Auto now consumes credits from your monthly allowance at these API rates :cursor
Why it feels sudden:
The August announcement stated changes would take effect “at your next billing cycle”, so monthly subscribers are only now experiencing the full impactcursor
Since Cursor 2.0 (late October), Auto has become significantly slower and appears to route more frequently to GPT-5 Codex, which many users find less efficient reddit+2
Longer response times and more tokens = faster credit burn cursor+1
The solution:
Stop using Auto and manually select a specific model like Claude Sonnet or Composer 1. This gives you much better control over your usage, and for most coding tasks, these models perform just as well (or better) than the current Auto behavior without the unpredictable credit consumption.cursor+1
Many in the community have switched to manual model selection and report much more stable, predictable usage.
+1 to this. The current limits are untenable with the rate of consumption for how I use Cursor and to be honest, this would push me to cancel and go with a competitor.
