Today, I had some time to think about something.
At first, I was a bit upset to hear that the AUTO model is going to cost money. For a moment, it felt like the last “free” thing I had was being taken away.
But then I asked myself: Would I really want to go back to the old pricing model??, where everything was billed per request?
No. Absolutely not.
Just look at all the other code editors that charge per request. Their models are constantly trying to generate the smallest amount of code possible just to save the company money. The quality suffers.
Even Cursor, since switching to the new pricing model, has introduced things like TODO mode, and the models have become noticeably less lazy.
As a paid developer, the new pricing while annoying at first actually saves me more time (and therefore makes me more money) than it costs. The efficiency gains from Cursor’s improved output far outweigh the subscription fee.
And here’s the key:
The new pricing gives the Cursor team the freedom to truly focus on making the models work as well as possible.
Bigger context windows
If a company charges per request, it will almost always try to keep the context window as small as possible to cut costs. Go ahead, try to find another company offering a 1M token context window for any model besides Cursor. You won’t. No company that charges per request can afford it [I might be wrong about this specific detail this is just as far as I’ve been able to check and find out]
Bottom line:
It’s easy to complain, and yes, we all like free stuff. But for a professional developer not as a hobby Cursor’s current pricing model is far better than the old per-request system.
I think this could also be the reason why, in the end, everyone comes back to Cursor.
I cannot comment on that precisely as my trial was eaten in 3 hours by auto mode. By that I would prefer a request-based system (I do not know how it was; but I surely did not make 500 requests in 3 hours).
But that’s not the point the point is that with five hundred requests, you’d get less quality. What you want, at the end of the process, is to save yourself as much time as possible, and in that time get a working piece of software without bugs or security issues, without having to keep going back to the agent to request fixes.
For me that is the point. Different perspective. But as I see, very little is happy for the changes. I read a lot the forum; it is an unsustainable business model imho.
This is the part that always interests me:
How many users does Cursor actually have?
And how many of them are the ones who aren’t satisfied?
Since there aren’t that many users here in the forum, it’s possible that those who are satisfied simply don’t bother to express their opinion they just use it, that’s all.
The ones who do come to express their opinion are the hobbyist users who have relatively more time, or people like me who enjoy chatting while their agent is working.
And hobbyist users usually have the smallest budget, care the most about cost, and care the least about things like security or perfect code, because they don’t know enough to notice it.
That’s what I’m afraid of though I might be wrong, of course.
That is a viable approach, but I am working between 2 replies here, too. Just not in cursor.
Reading the forum and mining some data many migrated, some were banned.
Money comes to play when you understand what you got for your money.
Read here:
When I am pissef off by a tool, I like to waste my time on making critics; maybe devs will improve it, but it is very rare, That is a kind of venting. Human attitude.
The 500 requests and then usage pricing (just the same cost per request depending on model) was incredible. I spent $200 in 1 month and got over 3.5 million lines of code. Not sure what you are talking about AT ALL when you are saying the requests gave you minimal output. Perhaps you were laser focused and only looking to make targeted changes with minimal code output, then I can see your point but frankly, I’d say Cursor wasn’t for you.
I mean that respectfully. I think they could have enabled a pricing toggle option to change pricing for use cases as such, but I believe Cursor was made to do significant work at scale which is where request based pricing was unbeatable.
In practice, I think the only option is the one I suggested deals with the providers of a specific model. But what you’re suggesting, something more general, I don’t believe will work. Anthropic is a tough company and not really willing to compromise; they know that right now they still have no real competition.
This is true, and typically the vast majority of users never bother with forums. If they’re happy, they never say. If they’re unhappy, they just leave without saying.
I think the one who will win is a code editor that has its own models and they will be good models and it will host them itself. A company that does that will be unlimited in resources and will be able to offer competitive packages.