lol
It sounds like their âspecial sauceâ in the background between your prompt and anthropicâs API is just using claude calls to âimproveâ your prompt and now they have decided to charge you for that privilege. One AI loop (which happens way too frequently where it outputs the same function over and over) and youre usage will dry up before your eyes.
@danperks So will my annual fee for my August 2025 subscription also be limited?
This will result in the loss of many users, right
Does Claude Code integrate into VSCode to give you that line by line diff of what it changed where you can accept/deny the changes. Also are you able to copy paste or identify files in the request to Claude Code the context. Also does CC also create restore points with every request so you can undo its changes, maybe several requests back. These are the key features I need.
When that happens, I force it to stop, and then I check the usage on the web and it says it was not charged because of Error, but yea it will be like 10x as many tokens. So I think Cursor has a way of trying identify when that happens and not charge you.
If you have an annual that renews after September 15th, it wonât be limited until the next renewal. I think he has made that clear many times. Now will Cursor honor that like they honored other promises.
Iâm using an Ultra Plan, but after seeing this article, I canceled my subscription.
2 Likes
@li_chunsheng Annual subscriptions started before 15 September keep the current plan features including Auto for the duration of that subscription period and only revert to new plan on next renewal.
e.g. if subscription started on 3rd August 2025, the same plan would remain until 3rd August 2026 and then change to new structure.
1 Like
For checkpoints I never checked since I am using git for that since there is git, but will check. Diffs and file/context selection are, for my taste, nicer than in Cursor (github style, one by one, step by step).
Would you mind describing your workflow with Claude Code in VSCode? I am a bit confused how it is similar and different than Cursor. I am assuming you make a request in the terminal or via some extensions, it makes the changes, and then you can view a sort of Diff of all the changes but not choose certain parts to accept/deny like in Cursor, or can you?
Every tutorial I see of people using Claude Code, no one is programming. I get thatâs the goal, but letâs be honest, for people who can program that is a huge disadvantage if a tool is designed to basically keep you out of the code. We are not at a level of AI competency where it can 100% do everything I tell it to nor can I afford Opus to do everything, even if it could. Thatâs why I am looking for some genuine feedback for people who use Claude Code when they use to use Cursor.
1 Like
Actually, I like the workflow much more tbh.
Basically, it is command line tool which integrates with VS Code. So either you start it via extension (which just opens the terminal and types claude for you) or you do it yourself.
Anyway, the way I use it (and I am software developer) - I start with creating to do. This is for me the most amazing part. Sometimes I donât even go into the code itself, just get that aha moment with todo and then decide what I am gonna do myself, and what makes sense to do with CC. Creating to do is not one single prompt or something, you need to feed it specific documentation, files, discuss what you want to try to see what it will come with. I also wrote bunch of architectural tests (not readmes but tests) so I ask it always to pay the attention to it as whatever it comes with will have to pass those tests too.
And then if I decide to go into it implementing something, I am in âapprove every actionâ mode, so it basically starts executing the to do, and for every single step there is diff like in Github which I am reviewing and trying to stay on top. It is good because you can stop it in the right moment (for example, coupling in a wrong place, doing it the way you donât like, using the library you donât want to) and ask it to change that specific implementation in specific way, and it can do it without loosing the focus. Once the step is done, you go further.
This is the main reason I donât believe that in CC you can âspendâ your allowance or whatever, because if you have one head, you are reading what it is doing, you are actually planning and designing your stuff⌠I donât see how it is physically possible. And I feel I achieve much more quality with it.
I am not using Opus at all, I donât need it. Code is something I wrote, designed, modularised, and contexts are quite small. But I did pull recently quite big refactoring with it (wanted to change some architecture I did back at a time for the ease of maintenance) and it was huge help (following the process from above).
Another thing I love is context compaction: you see always where you are with the context, and if you are close and didnât finish what you wanted, for me this is the sign I screwed up somewhere on design, so it is clear git reset HEAD âhard and go back to the drawing board.
Btw. I also use local LLMs recently a lot, thatâs probably another reason I am using CC only for heavy lifting, but thatâs a separate discussion.
1 Like
@danperks - When you introduced new pricing recently for individuals you advertised that Auto is not charged - it lasted a month. Could you guys not strategize a bit before you make an announcement? I have only realized that you charge for Auto when I looked at my Team account where I have multiple developers working on.
Working wth you guys become like tossing of a coin. One day someone will mention Cursor in the Customer Service Handbook as an example of âHow to NOT deal with customersâ.
7 Likes
@danperks There is a lot of competition out there, github copilot is giving unlimited request of gpt 4.1 and gpt 5 mini. You are shooting yourself in the foot, donât do this.
1 Like
Heâs not. Cursor is a superior product. And besides Cursor just gave away unlimited gpt-5-high-fast request for a week. I personally got like $70 of free requests, and I am not a heavy user. I assume some got hundreds. Every company is probably getting deals with GPT right now. So github will probably pull their unlimited models at some point, or like Cursor offer unlimited free models to encourage early adoption of their product as a sort of marketing campaign to lower the bar of entry into using AI.
Also gpt-5-mini is dirt cheap with Cursor, it might as well be free. I think Cursor is trying to get to somewhere stable, and that is token pricing. Unlimited requests and per request costs are just not sustainable.
Companies at some point gain nothing by offering unlimited models for $20/mo. Obviously they at is going to get abused or something. I mean what if it cost cursor more than $20 to make the requests. You guys are just asking for free things that done add up. Other companies are just doing it temporarily, just like what Cursor did.
3 Likes
@danperks
Many others have already pointed this out, but I also believe that completely removing any form of unlimited usage is a very negative move for the product.
I personally convinced my CEO to adopt Cursor company-wide, and the main reason this was successful was because unlimited usage was available. Even if itâs with limitationsâsuch as a rate-limited model like ClaudeCode, or only allowing unlimited usage with a lighter model similar to CopilotâI strongly believe that some form of unlimited usage should remain.
I really like Cursor and want to continue using it, but with the new pricing model, my boss will not agree. As a result, itâs very likely that my company will cancel Cursor and move to using Copilot and ClaudeCode instead.
Now convince your CEO to pay .02 per request with GPT-5-mini if he wants âfreeâ. Thatâs 1k requests a month for your $16 monthly cost. Maybe your CEO should be more focused on what the better performing product that inevitably will earn him thousands of dollars in returns than saving a few dollars with âfreeâ requests. Nothing is free.
Yes cursor is a superior product if you are a noob â â â â vide coder. Superior my â â â .
Cursor is superior if you need an assistant AI while you code. Iâve tried other tools and Cursor is the most geared to actively programming, not just spamming prompts. But you seem like you have it out for Cursor, but if you can, try to explain why its for ânoobsâ and not superior to windsurf, trae, claude code, and copilot as a tool for developers who are intimately involved in the actual coding process.
An engineer doesnât need to be spoon-fed they already know how to code and solve problems. A skilled programmer uses AI as a tool to automate repetitive tasks and accelerate development, not as a substitute for their own expertise. Instead of asking AI to âbuild this or that,â they leverage it to work smarter and faster. I think you got your answer why cursor is superior to ânoobâ.