@BrettCooper4702
1:21 it's not an IED it's a IDE. omg such an explosive mistake.
@EduardsRuzga
You came to same conclusions I did in my video and blog. I called video Vibe Coding olympics. I am creator of MCP called Desktop Commnader that allows Claude Desktop to code. And what I found and hundreds of other users reported was that it works better for certain types of work then Cursor/Windsurf To try and show that I made a test run of 7 tools on same prompt measuring costs, user involvement and quality of result. To my discomfort Desktop Commander with Claude Desktop came on top edging out even Claude Code(by a little) And list of reasons I can think of is: - better pricing for Claude Pro/Max - that they waste way more tokens then Cursor/Windsurf but get better resuslts from it - that in Windsurf/Cursor you pay for messages but in Claude/ChatGPT they are spreading compute over the user base so you can get more then you pay for - And prompts in Claude are way better then in Windsurf/Cursor, those are very focused on coding work. That makes them conservative and not creative in comparison of what Claude pulls of All in all it echoes what you share here. Great video man. You got my subscribe.
@murattosundan
claude code doesn't seem to use thinking mode. I fixed several issues with claude sonnet 3.7 thinking while claude code couldn't fix it by itself...
@maxziebell4013
In your calculations, you forgot that Cursor has many people paying for the service but not using the full token contingent. So, there has to be an offset. I think there is a significant number of people who just want to be part of the Cursor gang, using it only occasionally, but are not power users. It would be interesting to get some statistics on that. But there surely is this kind of user in most subscription models.
@TheKimshot
Very insightful
@BeInnovativeToday
Nice, another point is the cost of processing is much lower for Claude in the supply chain (1. cloud ai host) profit markup -> ( 2. Claude) - profit markup -> (3. cursor ) -> profit markup -> (4. end customer)
@Diego-h4g2l
At one point back last year I gave one of this "all in one" ai subscription a go, since they seem a great deal. But I quickly realized while trying to create a landing page that the results sonnet 3 gave in anthropic's chat interface were substantially better. Later down the road I tried cursor, windsurf, etc. Same situation, claude Desktop gave better result, clearly a way better understanding of the project. Now I'm playing with firebase, which is being quite good actually for free , so I dropped both my claude pro subscription..... But I still need to resume a more complex project, which I think will be something I'll go with claude code on the max plan.
@RandomCodeProgram
Claude Code opens many more usage scenarios compared to Cursor, etc. I use it to work with Obsidian; it's helpful for DevOps, particularly for writing and updating configuration files. It can be used in parallel with any other IDE (yes, without fancy integration, but I don't care). That said, I don't think everyone is comfortable with the terminal window, which is why those VS Code forks are thriving.
@taleem.help1
they are not buying at this rate . these are not corporate rates
@ephimp3189
realistically, with such wide user base in Cursor, there will be a significant percentage of users who barely use it at all. There is probably only a small percentage of users that actually hit the monthly caps. We don't know the exact numbers, but it's a safe to say the majority of users aren't hitting their monthly caps. Recent versions of cursor also have the model-auto select feature, again, the majority of casual users aren't going to change that to claude, so they probably user cheaper models
@pepa007
One unfortunate effect of this is that you tend to blame models for being "stupid". When in fact it's often the processing by the tool standing between you and the model. So sometimes you are using Claude via Cursor/Windsurf/etc. and keep running in circles. When you do, you just need to fire up Claude Code and it will sort out the same problem in no time, without even asking too much from you.
@MontyCraig
Claude Code with Max subscription is the way!
@vasiovasio
Really useful overview, Greg! Working with Claude Code is a... how to say it, it is a little bit like blindly in comparison working with something integrated in the IDE itself. The CLI do Stuff, show some of them, but it is definetly Different Paradigm from the usual programming process. Do you have an IDE editor open in the background mode to check things from time to time? Another question - how does Cloud Coe react if you change something in the code? It is scanning the current state every time, or has some internal memory, and this results in errors and confusion for the LLM?
@yujiang99
This cursory review of Cursor is horrendously wrong in two major aspects. First, Cursor relies heavily on its own models. Two numbers to highlight here: they deployed more than 10 of their own models in production and the number of hits on their own models is 100 million per day. That's a big moat! Secondly, large foundational models are on their way to become semi-commodities. It's totally conceivable for Cursor to finetune Llama or DeepSeek base model for coding. Claude might be the best model for coding right now, but the margin is not that big. Why the OP is so wrong? He doesn't know how the sausage is being made at Cursor!
@beans9632
so if you have the max plan, do you still get charged for every input with the api tokens or will it just use up whatever is the limit for the max plan?
@awakenwithoutcoffee
why choose when you can run Claude Code inside Cursor :P
@StrykEdits
With Claude 4 releasing and having VS code integration, should I drop Cursor? I pay roughly $100 a month to use it for daily coding. I tried Claude code when it came out and found it insanely expensive and inaccurate - mind you this was the early release version and prior to the Max plan.
@muray82
What you forgot in cursor is the Enterprise model (and rest of the tools). It will cost $40 while in theory the amount of request in the plan is the same. It's not the average Joe that is paying for the excessive tokens it's the enterprise. It's enough you have architect that have license but uses it once per week or 2. For such company it's a win. An employee on a sick leave - a win. Your calculations ignore that fact completely that Big Tech - even Antropic is paying for Cursor. Take such company as Microsoft where you have thousands of employees multiply it by $40 then assume 10% is not using it weekly. Another 30% is not a heavy user - your calculations will average out average Joe.
@brunonoleto8855
I never tested Claude Code, but for me there's no better cost-benefit deal of using Cursor with Gemini 2.5 where the slow requests are fast and unlimited and if you are able to manage well each chat.
@gregf123