I’d like to express that Cursor has become too expensive. It’s a great tool, but the new pricing makes it hard for many independent developers or those from regions with lower purchasing power to afford it. Please consider more affordable or region-based pricing plans.
13 Likes
Hey, sorry to hear you are unhappy with our new pricing. Unfortunately, as models have improved, they have also gotten better at reading more of your code, and writing more code in response.
This meant the old, 500 request usage plan wasn’t sustainable for the future! Which each request is more expensive to execute, the value the user gets is much higher too.
To reflect this, we switched to a simpler system, now giving you a $ amount of usage each month with is at least the same as your plan amount, but most likely more.
This does mean that heavy users of the most expensive models, like Claude 4 Sonnet, may get less than 500 requests each month. However, you can also now get more than 500 requests in a month from cheaper models, including Gemini 2.5 Pro!
In v1.4 which we just released, we’ve added an in-app tracker for your usage to help you know how much you are consuming as you work!
The best recommendation I can give is to give other models and configurations a try. It can be tempting to always use a model with thinking mode enabled, or to always use the best models. However, we’ve found that the vast majority of requests to models in Cursor could’ve been answered by a cheaper model just as well, if not quicker!
We are continuing to work on efficiency to hopefully make things even better in the future, but I do apologise if we have worsened things for you with this change!
6 Likes
In Claude Code, It only generate meaningful content, while Cursor produces a lot of redundant information that wastes tokens
2 Likes
I experienced this too, I subed a pro $20 on Aug 5th i.e, two days ago, to my utter amazement, I am told that I have hit my limit today! This is the worst I’ve experienced.
2 Likes
Unfortunately, as models have improved, they have also gotten better at reading more of your code, and writing more code in response.
so the pricing model was perfectly fine for older models? but newer models need more tokens and cost more? how come older Claude versions still costs the same as the newer models?
This meant the old, 500 request usage plan wasn’t sustainable for the future! Which each request is more expensive to execute, the value the user gets is much higher too.
dependending on how you calculate, the new pricing model is providing about 1/10 of the value compared with the old one. this is not due to improved model capabilities but graceless, aggressive market strategy.
This does mean that heavy users of the most expensive models, like Claude 4 Sonnet, may get less than 500 requests each month. However, you can also now get more than 500 requests in a month from cheaper models, including Gemini 2.5 Pro!
… not only does this imply that EVERYONE is a heavy user now (review your forum), but also this is quite wrong for the gemini 2.5 pro case.
In v1.4 which we just released, we’ve added an in-app tracker for your usage to help you know how much you are consuming as you work!
your usage meter is not correctly updating. neither during normal use, nor after restarting cursor IDE.
3 Likes
You guys always complain about high prices but don’t find a solution? You can chat via mcp server
I got solutions. I just keep stumbling upon the way, how s*** is being sold as gold.
1 Like
You should switch to Claude Code—it costs the same, but you can use it for much longer overall.
I don’t understand why Cursor consumes so many tokens compared with Claude Code.
Unfortunately, as the models have improved, they have also become better at reading more of your code and generating longer responses.
—
That’s not true: even with Claude 3.5 you can’t make 500 requests now. In the past, you could still use it after 500 requests, just at a slower speed.
Claude Code also uses Claude Sonnet 4.0, but it consumes far fewer tokens.
1 Like
The new models cost more—fine. But I was perfectly fine with spreading my usage between 1x and 2x request counts, as before. In 80% of cases, I would use a 1x request model, and only in very specific cases would I switch to a 2x model.
I knew exactly how much I spent, how much remained, and could choose models based on a fixed unit: the request.
The system now is definitely worse. Some new models cost more? Then introduce 3x and 4x request counts. At least that way, we remain fully in control.
1 Like
today I got the refill of cursors. Okay, so I start my day with my fast chat to review an implementation. How much I covered the implementation and all I created plans with agent mode and I used Sonet 4. Thinking, guess what? Within 4 to 5 requests, I drained 50 percent of the full month quota. How someone can survive this in a full month? I can do all these manually, but it took time, right ?? this i s what someone is paying for, right? this is litterally looting i can say
1 Like
I have to laugh at the complaints, y’all fed the beast (Cursor) and now it’s eating your lunch.
Tons of other options out there that are just as good that cost less. No reason to stay with cursor.
2 Likes
claude-4-sonnet is one of the most expensive models available
I would advise using gemini-flash
So what I’ve found useful for me is multiple subscriptions. I have 2 cursor Pro subscriptions (1 Yearly sub and 1 Student account I got for free) then I have a Claude Code sub, and I use Gemini CLI when all other limits have been hit. The way my work flow goes is I use Claude Code which has a limit that refreshes every 5 hours and it usually lasts me about 2 hours. Then when I hit my limit I will use Cursor Chat (usually o3 to use less requests). If I’m doing something easy I will use Gemini CLI, but I typically find it to be less reliable and I don’t trust it to do complicated tasks (but it does have the most generous rate limits). Using this combination allows me stay working pretty consistently without having to pay $100 - 200 a month.
I would really like Cursor to switch to something like Claude Code’s 5 hour rate limit refreshes. My cursor accounts are pretty much dead after I hit my rate limits because the direct API costs are too expensive for a solo dev that’s in the middle of production.
One loophole I have found is after giving a fairly complex task, the model will inevitably miss something. In the past, I would stop and redirect it which saved compute time for Cursor (and cost me an extra token.)
Now, I just let it run, which probably uses more compute time but saves me a token and the model eventually figures out what it missed. It’s less efficient but cheaper.
No, Claude is expensive. Do not use Claude models.
Oh man!! I just used cursor for like an hour and it cost me over $17!!! I can get a real human at these rates. I’m sorry Cursor folks but I don’t think I can stay in much longer.
1 Like
You’ve got it wrong. You can’t blame Cursor for the price increase. It’s because the price of the model has also gone up.
When the river rises, the boat floats.
To me the issue isn’t so much the fact that models like Claude 4 consumes so much (I mean it is a problem, but I can somewhat understand where it comes from), but it’s the fact that auto mode is basically useless meaning that once you have reached your limit, what remains is a glorified VS Code until the limit resets.
Give us 500 requests back!
I used for 2 f****** days and almost hit limit! That’s nonsense, before I use entire month and never hit the 500 requests.
And, yes we can blame cursor, because models like sonnet have no price changes.
Last month was unlimited, now I pay $20 for 2 day use.
3 Likes
anthropic did not changed price for sonnet