make it free for 1 week like gpt 5 pls:))))
3 Likes
I’d highly suggest testing other apps using the new Claude Sonnet 4.5 model. You need to explore a lot of the BS out there, but honestly you might find something special hiding in plain sight. Cursor charges heavy on usage, while (01000111 01101001 01110100 01001000 01110101 01100010 00100000 01000011 01101111 01110000 01101001 01101100 01101111 01110100) runs at 1× cost multiplier (the base rate, no inflated usage charge). 
But, Cursor is the best sleek application to date. 
this model creates a lof of noise you have to pay for, even if you ask don’t do some particulat thing it does. I have impression that developers of this model just want to generate as much strings as possible to charge the users. useless.
2 Likes
Interesting!
I take back what I said guys, claude 4.5 is far superior than gpt 5 high. Solved bugs that gpt was not able to solve, it’s drastically faster and benchmarks show that it has 77.2% SWE vs 74.9% gpt 5 high.
I am using claude sonnet 4.5 for now.
3 Likes
I’m still in the legacy 500-request plan, and I see it’s consuming more than one request per prompt. First it consumes one request, but then it consumes more, not sure if it’s a tool use thing, or a tokens thing. But Sonnet 4 consumed exactly 1 request per prompt. This makes it hard to keep track, and feels buggy (I used 1 prompt, but I was charged multiple, and this happened multiple times)
1 Like
hi @Raz when you go to Dashboard and then menu Usage, you will see a detailed breakdown for each request with the cost.
Have you used the Thinking version of Sonnet 4.5?
Also used thinking with max enabled (by accident). Tried browser use and it worked pretty nicely, but now comes the bummer… 
It ended up burning like shitloads of tokens with every requests, so I tell you, rather be safe checking your spend regularly, than sorry 
I assume though that there is something really wrong with the caching. it doees not make sense at all, I was having like 20+ requests with 400k+ tokens and zero caching!
I think it’s also unrealistic to expect you can use Sonnet 4.5 for every prompt. You have already stated that some models are cheaper, so use those if they can accomplish the same task, and only if necessary use sonnet 4.5. It’s not Cursor, it’s Anthropic making the model expensive or making sonnet use a bunch of tokens.
It might be that sonnet 4.5 is more expensive, specifically when used in max and thinking mode. But how can it be that MANY, sorry, ALL subsequent calls on the exact same chat thread use ZERO caching and charge like $3+ dollars per prompt if the thread gets in the 500k token range and like 80+ premium requests. I never ever had this with GPT 5, there any follow ups normally do not go up to using like 2-3 requests even in MAX mode. That must be a bug! 
Claude equals tokenspam, ever since it came out. I rarely use it, sometimes to give a hint or diagnose as a second (or third) opinion. The optimism of the model responses annoys me a lot, which also covers up the tokenspam it produces.
4 Likes
Something is weird with the model pricing this request was costing $0.17.
to explain, cached should be 10%, read is 3$/1MTK, write is 15$/1MTK, it should have not costed $0.17.
to reiterate for those that don’t understand, only around 3,000 tokens without caching, the rest 65,000 tokens were cached, this is insane pricing or something is wrong with my understanding.
while gpt-5-fast which is 2x than gpt-5, the request cost $0.22:
can someone explain what is going on with 4.5 sonnet thinking pricing?
2 Likes
For me claude sonet creates a python script (while language of project is differrent) the only goal of the script to print very long message that things has been really succesfully done (when it was not)! then it repeated this message in response - all those things you have to pay for.
Hello, I don’t know if this is the place but I’m new and I’m asking for your help and assistance. I want to build an ERP system from scratch. It will be for a backend with Laravel and a frontend with Vue3. Since I’m not a programmer and I’m wondering when I pay a subscription to Cursor, will it be enough? I’m talking about the Pro Plus plan for $60 per month. Separately, I thought it would be a good idea to have Chat GPT 5 write the prompts, and so give them to Cursor for a better understanding of Cursor. When I pay for the Pro Plus plan, will it be enough for me to slowly build my system? Separately, if I also pay for Chat GPT 5 on a monthly basis, is it worth paying for both or just Cursor. I thought a lot about Claude, but when I read how it eats up tokens, I gave up a little. My idea is to ask you for help on what to use to make my first project. I don’t want to copy codes with different AIs, and if possible, everything in one place. Should it be GPT 5- Codex, should I do things in Cursor? I need help from you people who understand and have been on my first starting path. Thank you from the bottom of my heart!
60 usd will get you nowhere. if you want speed. you need at least 300 USD. 200 for ultra plan and 100 for extra pay as you go. Unless you want to use Claude Code Extension, which can give you more usage for 200 USD.
Claude 4.5 is coding for hours now. I don’t if I should leave the chair and go for walk or just sit here and watch it.
How do I understand this? If I want to work on the system module by module for 3.4 hours a day, $60 won’t get me anywhere? Do I have to use the biggest plan of $200 to do what I want? And do you think I should pay separately for cHAT GPT 5 or does it make sense? Is it better to pay per month only for Cursor?
120 hours a month costs you more than 60 USD.
What i see, is that models are getting better. So if you don’t do anything and wait 6 months, a model will emerge which might finish your product in a week. you can also start now and work everyday, 3-4 hour, which will sum up to 120 hours a month give or take, and I think you will use more than 100 usd for 120 hours. I work like 200 or a little more and I spent 220 last month.
Claude 4.5 is good example. it does better job in one go than its predecessor for the same cost of tokens, but it is using less tokens because of its efficacy.
So either wait and work on your specs and make them as sharp as possible and split them into smallest steps as possible, and when done, unleash 4.5 or by then Claude 5 on it.
We have entered SpecDriven development supported by real e2e test like playwright. mock tests and unit tests are not good anymore.
If you make sure your pipeline is flawless, you will get to your goal cheap and fast.
I’m thinking of starting somewhere in December during the holidays. My documentation is ready step by step and module by module, I just don’t know where to orient myself and which model to do it with. Whether with Claude 4.5 + Cursor or Cursor + GPT5. If I pay for the maximum Cursor plan, do I have to pay separately for Claude or not? Cursopr is easier for me as a beginner and I want everything to go through it. But I don’t know if I pay only for it will it be enough for me or do I have to pay separately for Claude in this case? What would you advise me as a professional?
I’m using grok-code-fast-1 for coding and GPT5-low for asking question . claude4.5 too expensive.