Just to update: it took more than 23 hours for my Claude 4 Sonnet to work again on Cursor.
This is really unacceptable for anyone who depends on the tool to work. We need a clear limit, ETA, or at least a countdown. It’s impossible to plan anything this way.
1 Like
Same here - my charges went way up. I was spending less than $100 before, and now I’ve already spent $60 in 4 days. This is insane. No warning, no transparency.
3 Likes
Please brother switch back to legacy system if you’re on the pro plan ($20). The new rates are completely irrational!
These are topics that were similarly hidden:
1 Like
I would love to, but the cursor team simply removed this option when I switched to pro+
This is worst than I thought if that’s the case. You don’t have the ability to switch back to legacy even though you’re paying 3x more? That’s absurd.
I have been emailing back and forth with Cursor on this topic since yesterday and this is the info I got back from Cursor:
The Pro plan includes extended agent limits based on compute usage that reset at the beginning of each month. When you reach these limits, you'll see the option to enable usage-based pricing (at API rates + 20% markup). This explains why you started seeing the message on July 1st - it was the monthly reset.
The Pro plan's extended agent limits are based on compute usage, not just request count. Looking at your screenshot, the July 1st request that triggered usage-based pricing had a significantly higher token count (633,541) compared to other requests. This means that single request used enough compute to exceed your included monthly allowance, triggering usage-based pricing.
Your subsequent requests show as "Included" because they used less compute and fell within your remaining monthly allowance. The system optimizes to use your included allowance when possible, which is why you see a mix of "Included" and "Usage-based" requests.
The Pro plan's rate limits are based on two components that reset and refill differently:
1. Burst rate limits: Allow for short periods of high usage but refill slowly
2. Local rate limits: Refill fully every few hours for regular usage
This explains why you can see some high-token requests marked as "Included" while lower ones are "Usage-based" - it depends on your current rate limit status at the exact moment of the request, not just the token count. When either limit is depleted, subsequent requests trigger usage-based pricing until the limits refill.
Ignore some of the numbers in the responses from Cursor as I was using my own history to illustrate my point. My last response to Cursor was that it would be great if they actually exposed the refill percentages on both the local rate limits and the burst rate limits. Based on these exchanges, without those, there is no way to truly monitor our usage and to avoid going from Included requests to Usage-based requests.
Very intresting…
People are reporting this is no longer calculating the same as it did before the changes. I cant confirm, although at one point I did switch back to check my own token usage.
It sounds like you would benefit by switching to the Pro+ plan
I just switched so I will check. I was pretty happy right before the change, Sonnet 4 (non-thinking) was .5 points per request. So at 1000 requests per month that worked well for me and was predictable.
I guess I will find out pretty quickly if the old system is not really the old system anymore.
I think the new system needs some work but I could live with it if there was a way to see how/when things were going to turn off access and when that happened how long until I get access back.
It’s totally reasonable that they want to stop the bleeding. $20 for the frontier model usage and super huge context is bonkers. But they just need to be transparent in a way the user can make smart decisions.
I think in a world where they throttle my access transparently I can moderate my use of Cursor with Claude Code (Pro) and Google CLI requests and hold out for Cursor when I am worried a change might be more likely to be a breaking one. But that’s only possible when I can see where the wall is so I know when I am going to hit it.
Also maybe free tool use (with low cost models or via OpenRouter free tool use models) and Claude Code / Google CLI MCP servers could be the answer. Like telling Cursor to use those options when the requests are less robust to avoid using Cursor compute or whatever it is that costs us. It’s hard to know cause everything is pretty cloak and dagger as to how the limits are set (and reset)
1 Like
I found this topic linked from another topic, I want to address some points:
- Legacy plan works as before, can’t confirm about usage-based pricing and can’t confirm if switching to Pro+ would disallow the option or if it will disappear in the near future, I hope not as we should choose the type of workflow we want.
- I can understand why this topic or others get hidden, it’s not about customer feedback but from some users behaviors suggesting other tools or that cursor rug pulled, trying to cause panic, remember these tactics are known and used by paid actors, even if you aren’t, forum admins cannot know.
- I agree reset timing clarity should be immediately implemented and if the discussion was about it this topic would havn’t been shadowbanned
- I would suggest cursor to do an internal investigation on who thought not having reset timing clarity was a good idea.
- To users telling X company gives unlimited requests: they’re stealing customers from cursor-like software by losing money but also acquiring all your codebase information, nothing is free, if you’re ok with that go forward.
Yep. I haven’t been able to use Sonnet in over 24 hours now. In the same time I have been using Claude-Code and pumping out a whole new business website. Do I hit a limit on Claude-Code? Yes. But it at least says “Reaching Usage Limit” and then “Usage will be reset at X” where x is time and it is true to its word. It HAS transparency at least.
Not an airport, I know. But I think if this keeps up will soon depart usage of Cursor.
3 Likes
We got an answer
1 Like
Hell yea!
According to the clarification we have:
We were not clear that “unlimited usage” was only for Auto and not all other models , which have at least $20 of included usage. Based on median token usage, the Pro plan currently covers about 225 Sonnet 4 requests, 550 Gemini requests, or 650 GPT 4.1 requests, and the vast majority of Pro users do not run out of their included usage.
And now lets do some math:
Claude Code Pro ($20/month):
- 45 messages every 5 hours
- 5-hour resets = ~4.8 resets per day
- 45 × 4.8 = ~216 messages per day
- 216 × 30 days = ~6,480 messages per month
Cursor Pro ($20/month):
- 225 Sonnet 4 requests per month total
The Comparison:
Claude Code: ~6,480 interactions/month
Cursor: 225 interactions/month
That’s a 29:1 ratio! 
The value proposition is so bad it’s almost comical. Hope they will not change or remove old pricing model.
6 Likes
agree with your point of view, and my alternative plan is already underway. The purpose of the cursor is to induce you to spend more unnecessary money and to train people to say: yes, this is the money you should have (the trick of the monopolistic capitalists). I do not agree with this behavior and firmly resist it. I am paying attention to the development of trae, and of course, if Claude Code is good, it will be the next option. 
3 Likes
I have canceled the renewal of my 1-year Pro plan. I’m extremely disappointed. Users’ issues are not being addressed satisfactorily. For those who pay for a service but receive such poor quality, it’s better to switch to another provider. There may be some initial confusion, but trust me, it’s a much better experience and feeling overall.
1 Like
Legacy plan is similar but when the plan switched from Legacy to this current plan Sonnet 4 uses (non-thinking) were showing up as 0.5 credits. Now each use of Sonnet 4 uses 1 credit from what I can tell on the legacy plan. I guess if the new plan provides 225 uses and if you think your uses are above the median then you definitely want to be on the legacy plan.
1 Like