More like $20 an hour. lol.
Yes I agree. I’m a vibe-coder, maybe do an app a month or every two months. I don’t use the “max” ones but when regular Claude upgrades, I point to that one, because I like it. The last one creates an md file as a result of every prompt and simple update to the code, and I believe this is where the credits go. I didn’t like working with “Auto” if I don’t see in real time which model is chosen (and I’m not sure how it works apart from trying to save money). Cursor warns me that in X days I’m projected to deplete my PRO and then I’ll have to wait.
I’m sorry, but I don’t understand this behavior. It didn’t use to be like this - there is no way a vibe coder who opens Cursor 1-2 times a month will have a PRO subscription depleted.
/Shmulik.
Lol bruh. Never use Anthropic models. Always use gpt5-mini or grok 4 fast. Then gpt5-high or codex for harder problems. Anthropic is overpriced and is for noobs.
Absolutely, but Sonnet does a great job with design.
Vscode copilot is 20$ for 500 premium agent requests. Free unlimited gpt5 mini.
You could do 2000 haiku 4.5 agents requests for 20$.
Interestingly: after paying $200, you got $504.
By the way, why didn’t you use gpt-5 much? Is it too slow?
You’re wrong.
I assume that caching works the same way, despite the difference on the dashboard. The thing is how Claude uses his tools: he does this very often and likes to make small edits; while gpt-5 can immediately send a request to read multiple files, search through multiple files, edit multiple files. And the file editing itself takes place in large chunks. This has a drastic effect on the number of cache reads. And besides that, gpt-5 also has twice as cheap tokens.
I saw the comments here that 16-20$ is not that much but what they don’t know that this is too much in other countries where we are not getting that much in return from cursor
Michael, they don’t care about us ![]()
Why are you using Cursor and not Windsurf or Trae then? Would you also complain that a Ferrari is not just 500k if you account for the maintenance costs and a Wolkswagen is way cheaper? It’s a valid complaint, but don’t drive around a Ferrari if you can’t afford one.
Do other tech services offer discounted rates to countries with lower average income? Also people in lower income/living expense countries can often compete for jobs with people in higher income/living expenses. It cost Cursor the same to provide a service to a user no matter what country they are from, and if other AI services (or tech services in general) are not giving discounts, then why would Cursor.
@rodat-dev Right, just use a cheaper product. Instead of asking for cheaper pricing but refuse to choose the cheaper services first.
Cursor is absolutely doing the right business moves, considering how much all the AI providers are known to be at a loss and dependent on constant influx of investor money. Staying out of this overinflated market where somehow the providers have to pretend that AI is low cost to grab a customer base is the only way to survive
And even then, Cursor is not really that much more expensive.
I would pay more, just not 4x more. Auto is useless and a decent model not lasting ½ a month is crazy. I can already use a ton a models directly in something like T3 chat can just copy code in / out as needed for $8 a month. $20 - $40 a month would be fine for Cursor if Auto uses a decent model.
T3 gets you 100 claude requests a month for $8? Pretty good deal I suppose, but wouldn’t it be more effective to just use Claude Code then? As for other models, if something like T3 works for you, and you want to copy paste code back and forth to save $20-30/mo, then I would question where your value is. For most professionals, $20-30/mo to even save 1-2 hours of work would already be worth it. Hard to imagine ever going back to copying and pasting code into a chat instead of using a tool like Cursor.
Auto is decent for simple things (if you know how to program, gradually build out the tasks, and word the requests), and GPT-5-mini does a bit better but slower and is free. Then there is Grok-Code which is pretty good, fast, but not great without babysitting it and giving it super clear instructions (free/very inexpensive). GPT-5 is probably the most well rounded for the cost, but Sonnet 4.5 is the best in my experience for complex tasks, but it does cost a ton and if people are using that for most of their requests on the $20 plan, yea it may last a day or two. But $20 is very little money for the work you are getting in my opinion. So with T3 you could get like 200 sonnet 4.5 requests I think, not 100% sure how it works since they are not clear on their website.
So use cheaper models or pay more money… or use something like T3 and save some money and take a hit with your efficiency. Or use Trae if you don’t need certain Cursor’s features. Not sure how long Trae will be this cheap, doesn’t seem sustainable, but who knows.
Mr. Cursor Advocate, keep calm. Not everyone must agree with your thinking
@MidnightOak Man, what are you? You’re either one of these two:
-
A PR guy for Cursor, or
-
Someone with way too much free time, working on basic HTML/CSS projects (where Auto works best), and spending every minute arguing on this forum.
Go get some real work instead of sitting here defending Cursor all day. We’re just sharing our concerns with the Cursor team , not with you. So please, take your debates elsewhere and stop flooding every thread with your comments. Let the Cursor team read and respond themselves. No one here needs you to speak on their behalf.
He quoted my joke and wrote a novel. crazy guy
You guys just want some one sided discussion. Not sure why you guys get annoyed when I don’t I don’t join your dogpile on cursor. I am providing my opinion just like you guys are providing yours. Just because I think Cursor is worth the $$ as of now, doesn’t mean my opinion no longer is valid or shouldn’t exist on the forum.
@alihamza79 I don’t think you know what a forum is. It’s not your private space to just complain about cursor. If you want to share your concerns with the cursor team privately then send an email or make a post and don’t read the replies at least. And obviously everybody uses cursor differently and there’s some people who are working on more complex projects and some that are simpler. Those that are simpler or where they do a lot of the manual programming themselves still can get away with a $20 plan or get a lot of benefit out of Auto. I still do a considerable amount of programming myself in the game development projects I’m working on, so Auto and the Pro plan are good enough. If that invalidates my opinion because it’s not up to your arbitrary standard then that’s on you.
That was my bad. Meant to reply to @Syed_Faraz_Shahid (right above your post) about the real cost of Cursor in poorer countries. My reply makes no sense in the context of your joke. So yea, that is crazy.