I would treat Auto as a routing convenience, not as a separate unlimited bucket.
The confusing part is that “Auto” sounds like a free mode, but in practice people still need visibility into what it routed to and what budget it consumed. Without that, it is hard to tell whether you hit a policy limit, a model specific cap, or just normal monthly usage.
Updates to Teams and Auto · Cursor I guess you could refer to this, and also note that on the first month of subscription, some people can get unlimited auto as like.. a showcase but no, unlimited auto stopped being a thing last year already
Hey, thanks for the question. Auto isn’t a separate unlimited plan, it’s just smart routing between models. Requests via Auto still use your Pro plan’s included monthly usage, same as the regular models. Auto just usually picks a cheaper or more efficient option. When your included usage runs out, access is blocked until the monthly reset, that’s what the banner in your screenshot is showing.
So this isn’t a new change just for you, that’s how the current pricing model works. Unlimited Auto existed earlier, and sometimes in the first month as a promo, but it isn’t the current policy.
What you can do:
Wait for the monthly limit reset.
Turn on on-demand usage via the Set new limit button. Then you can keep going past the included limit and pay as you go, and you can set your own spending cap.
I’ve been a heavy daily Cursor Pro user for many months and never hit limits. This month I incurred $62 in on-demand charges, and on the first day of my new cycle I’m already at 29% API usage using only Sonnet 4.6 Medium, with Auto off, ChatGPT for planning, and Ask mode instead of Agent. Has something changed in how usage is tracked or billed?
I would treat that as an account-specific usage question, not just an Auto question.
The important detail is that your pattern changed even with Auto off. If the dashboard can show daily usage or per-model breakdown, I’d compare the spike day by day first. Otherwise support is probably the only place that can confirm whether the accounting changed or whether one workflow started consuming much more context.
Hey @msterling10, there hasn’t been a broad change to how usage is tracked or billed that we’re announcing. The pricing model is the same as described above. If your numbers feel off compared to how you’ve actually been using Cursor, that’s account-specific and worth a direct look.
Account and billing questions are handled by our team at [email protected]. They can review your specific usage and charges.
Can’t thank enough for everything that has been given to me by cursor.
but the auto model is really great, now that I know it routes to the most effective/cheap model I will only be using this. I mean outside of cursor I was having a hard time chosing the best model for any task.
Last month, after reaching my limit, I was still able to use Auto. This month, Auto stops completely. Could you explain why the behavior is different? Has anything changed in the way Auto works or in the quota system?
Interestingly, your CEO mentioned in the updates from July 4 and August 12, 2025, that the “Auto” feature was indeed intended for “unlimited” use. The issue now is that the feature seems to become completely unusable once the quota is exhausted. While most users might not actually require truly unlimited access, the fact remains that we subscribe to your plans; yet, once the Auto quota pool is depleted, the feature becomes entirely unavailable. I’m not sure if this is standard practice for Cursor, but why are changes that significantly impact user decisions made quietly, without notice? Previously, there might have been access to a “slow pool” after the main Auto quota ran out. Is a Pro Pro+ Ultra user’s subscription essentially rendered useless for the remainder of the month just because they hit that “fixed quota”? That hardly seems reasonable.
Changes always happen quietly. For instance, the changes to the usage page—as discussed at https://forum.cursor.com/t/usage-page-to-token-amount-what/167153/69—reflect this trend. It is becoming increasingly like a “black box”: first, the concept of “unlimited auto” was made vague, and then support for slow requests was quietly removed.