What specific issue do you have with it? What mode do you use? What prompts do you use?
I think we will not wait alot to find a good alternative, companies couldnât compete with cursor before but not after their new pricing policy.
donât use Claude-4 
just use grok-4
good output input and no stupid cache over 2m per request
i do not use memories
Yes, same here. I ran out of my âallowed subscription usageâ in a matter of 2-3 days as well, without a clear explanation of why, and am still somehow charged additional costs on top of separate API calls due to running out of aforementioned subscription usage. I think a lot of us are lost on just how the pricing works.
I am trying it right now, it just created me a long plan for a simple problem that needs file reading, research and one code line fix
lol
yep, I set up spending cap of 10$, I exhausted them in 2 hours
Same issue of huge tokens usages starting 14th of July. Passing from less than $0.10 per request to $2
While memories are great to help the model have a longer-term understanding of your codebase, a lot of any context (memories, files, chat history, etc) can contribute to excessively high token usage.
In non-MAX mode, we limit the maximum context window to ensure this is minimised and requests are more predictable, but MAX requests (especially those with large context window models) have no cap, so can very high token usage surprisingly quickly.
To keep your token usage low, Iâd recommend not using MAX mode unless itâs necessary (which it often isnât!), routinely starting new conversations instead of continuing long ones on different topics, and trying to keep rules concise and to the point, as they are always included in every request you make!
The usage is INSANE. In the beginning I could code 16 hours a day for a month and only use around $50. Now we are half way through the month and I have spend $220..
Yeah, the token burn on Ultra is brutal, especially with Auto mode picking the priciest models. Quick tip: if youâre using Claude Sonnet 4 and not bringing your own API key, Cursor adds a 20% markup. You can avoid that by plugging in your own Anthropic key in settings. Same models, way cheaper usage. Helped me stretch my quota a lot further.
If you're using Claude Sonnet 4 via Cursor AI, you're probably paying 20% more than you need to
Yes, my usage increased also suspecially strong during the last 3 daysâŚ
Iâm out. This is ridiculous. Unusable.
Oh, I love how thatâs honest and up front on their page. Yep - I already cancelled. Cursor has been an absolutely awful and very expensive ride for me. The pricing is ridiculous and so is the usage.
They should offer full refunds for this month - itâs absolutely outrageous.
Hey all, just to clarify @0xHACKSâs message, Auto is actually free and unlimited on every plan, so does not have any effect on your usage!
Also, Iâd recommend people to check Cursor - The AI Code Editor for how much their usage wouldâve actually cost if you were using an API key direct with the model provider!
Ah, I see, appreciate the âclarification,â but that phrasing was deliberate. It caught attention, didnât it?
Now I also see why youâre avoiding being upfront about the real markup Cursor adds when people donât use their own API keys. Thatâs the actual issue, not just Auto mode.
Telling users âAuto is freeâ while skimming an extra 20% on Claude usage through your default config isnât exactly transparent. People deserve to know what theyâre really paying for, not just what sounds free on the surface.
Letâs not dress it up. Cursorâs a solid tool, but letâs be real about the business model.
If you're using Claude Sonnet 4 via Cursor AI, you're probably paying 20% more than you need to
Auto succeeds at simple to moderate tasks about 1 in every 20-30 prompts. Often times it goes on vision quests, hallucinates often. I donât find it useful at all. Context seems to be lost constantly.
Yeah using ultra it went up a lot? More or less x10 tokens 
Me too.Time to quit.
