光标切换在您的计划限制结束时按每个令牌收费。称为“按需使用”。
我曾是Cursor Pro的订阅用户。在1月14日,我达到了订阅使用限制。没有任何警告,也没有“嘿,你已经用完了包含的配额——想继续按每个token的费用使用吗?”这样的提示。Cursor就这样……继续运行。默默地将我切换到了他们所称的“按需”计费——这意味着从那时起我使用的每一个token都按照API的价格收费,而我对此毫不知情。
“按需使用”——谁会将其理解为后付费收费呢?
让我感到困惑的是。我在美国生活了多年,英语也没问题。但当我在账户中看到“按需使用”时,我真的以为这意味着在我的订阅计划内的使用——也就是说,我在需要时随时使用。就像按需流媒体服务一样。按需服务。在其他所有上下文中,这个短语都是这个意思。
但在这里并不是这样。在Cursor的世界里,“按需”意味着“你现在按每个token的全API价格收费,我们稍后会给你账单。”如果这不是故意误导的术语,至少也是糟糕的用户体验设计。
我在4天内不知不觉花掉了20美元
claude-4.5-opus-high-thinking模型每次请求的费用在0.50到4.00美元以上。我不知道我已经按token计费,所以我像往常一样继续使用。四天后,20美元就没了。
当我最终被阻止时,我以为我的订阅限制只是稍晚到期。用户界面提示我“添加API使用”,费用为20美元,所以我就加了——以为这是我可以提取的余额。
结果并不是。Cursor的客服后来告诉我:“这些不是预付款或充值——它们是已经发生的API使用费用。”
所以我并不是在添加信用。我是在提高一个已经默默累积的费用上限。用户界面对此没有任何提示。
账单
项目金额
Pro订阅 ~20美元
按需收费 #1(1月18日) 20美元
按需收费 #2(2月7日) 20美元
总计 ~2.5周 ~60美元
最终发票显示总的按需使用费用为42.12美元。在扣除第一次20美元的付款和因超出硬限制而获得的2.12美元退款后,我又被收取了20美元。
客服让事情变得更糟
我发邮件请求退款。被拒绝。好吧——我用了tokens,我接受这个。
但我无法接受的是:客服对费用的描述不准确。他们告诉我2月7日的费用是“17次对gpt-5.1-codex-max的调用,总计0.29美元”,并且“应用了20美元的最低收费。”
这让人感觉我为29美分的使用被收了20美元。实际上根本不是这样。20美元是42.12美元的剩余余额,涉及多个模型。为什么要这样表述?要么客服不理解自己的计费,要么他们试图用误导性的解释来阻止我的退款请求。
Cursor需要改进的地方
- 当达到订阅限制时,立即停止服务。不要默默切换到按token计费。要询问用户,获得明确同意。这是基本的。
- 更改“按需使用”的名称。没有人会将其理解为“后付费按token收费”。称其为“按使用计费”或“超额费用”。要诚实。
- 让“添加API信用”真正像信用一样工作。如果我点击一个按钮说我在添加20美元,我期待的是20美元的余额,而不是对我不知道的费用的默默提高支出上限。
- 培训客服准确解释计费。不要挑选一项费用让20美元的收费看起来像是最低费用问题,而实际上它是42美元总额的一部分。
我对Cursor用户的建议
- 不断检查你的账单页面。订阅到按需的切换是隐形的。
- 除非你在积极监控成本,否则避免使用Opus和高思维模型。一场会话可能花费10美元以上。
- 当你看到“添加API使用”时,要明白你并不是在添加余额。你是在提高支出上限。
- 如果你取消,注意几周后可能出现的费用。
我对Cursor已经失去信心。这个工具本身还不错,但计费系统似乎是为了通过混淆而非透明度来最大化收入。
还有其他人也遇到过这种情况吗?我真心想知道这是否是一个普遍问题,还是我只是个幸运儿。
查看原文
I was a Cursor Pro subscriber. On January 14th, I hit my subscription usage limit. No warning. No "Hey, you've used up your included quota — want to keep going at per-token rates?"
Cursor just... kept going. Silently switched me to what they call "On-Demand" billing — meaning every single token I used from that point was billed at API rates. And I had no idea.
"On-Demand usage" — who interprets that as post-paid charges?
Here's what gets me. I've lived in the US for years. My English is fine. But when I saw "On-Demand usage" in my account, I genuinely thought it meant usage within my subscription plan — as in, I'm using it on demand, whenever I need it. You know, like on-demand streaming. On-demand services. That's what the phrase means in literally every other context.
It does not mean that here. In Cursor's world, "On-Demand" means "you are now being charged per token at full API pricing and we will bill you later." If that's not deliberately misleading terminology, it's at minimum terrible UX design.
How I blew through $20 in 4 days without knowing
The claude-4.5-opus-high-thinking model costs $0.50–$4.00+ per request. I didn't know I was on per-token billing, so I kept using it like normal. Four days. $20 gone.
When I was finally blocked, I thought my subscription limit had just run out late. The UI prompted me to "add API usage" for $20, so I did — thinking it was a top-up balance I could draw from.
Nope. Cursor support later told me:<p>"These aren't prepayments or top-ups — they're charges for API usage that already happened."<p>So I wasn't adding credit. I was raising a spending cap on charges that had already been silently accumulating. The UI gave me zero indication of this.
The bill
ItemAmountPro subscription~$20On-Demand charge #1 (Jan 18)$20On-Demand charge #2 (Feb 7)$20Total for ~2.5 weeks~$60
The final invoice showed $42.12 in total On-Demand usage. After subtracting the first $20 payment and a $2.12 refund for exceeding the hard limit, I was charged another $20.
Support made it worse
I emailed asking for a refund. Denied. Fine — I used the tokens, I accept that.
But here's what I can't accept: support misrepresented the charges. They told me the Feb 7th charge was for "17 calls to gpt-5.1-codex-max totalling $0.29" with a "$20 minimum charge applied."
That made it sound like I was charged $20 for 29 cents of usage. That's not what happened at all. The $20 was the remaining balance of $42.12 across multiple models. Why frame it that way? Either support doesn't understand their own billing, or they were trying to shut down my refund request with a misleading explanation.
What Cursor needs to fix<p>Hard-stop when subscription limit is reached. Don't silently switch to per-token billing. Ask the user. Get explicit consent. This is basic.
Rename "On-Demand usage." Nobody interprets this as "post-paid per-token charges." Call it what it is: "Pay-per-use billing" or "Overage charges." Be honest.
Make "Add API credit" actually work like credit. If I click a button that says I'm adding $20, I expect a $20 balance. Not a silent spending cap increase on charges I didn't know existed.
Train support to explain billing accurately. Don't cherry-pick one line item to make a $20 charge look like a minimum fee issue when it's actually part of a $42 total.<p>My advice to Cursor users<p>Check your billing page constantly. The subscription-to-on-demand switch is invisible.
Avoid Opus and high-thinking models unless you're actively monitoring costs. One session can cost $10+.
When you see "add API usage," understand you're NOT adding a balance. You're raising a spending limit.
If you cancel, watch for charges that show up weeks later.<p>I'm done with Cursor. The tool itself is fine, but the billing system feels designed to extract maximum revenue through confusion rather than transparency.<p>Anyone else get caught by this? Genuinely curious if this is a widespread issue or if I'm just the lucky one.