Not sure it is a common thing:
In Cursor, compared with similar or same models in Cline or VSCode, the output often stopped after one or two tool calls.
Is this a recent change to save your cost?
I’m pretty regret to purchase a Pro version of Cursor in this case.
Same here, even max mode in gemini and claude sometimes terminates output without implementation. It is also counted as a success request.
now im using v0.48.9
It’s almost unusable now, often stopping after saying “Okay, I will modify ■■ document.”
It’s just a waste of points
A lot of people comment on problems. It would be nice if the cursor team could be a little clearer and more transparent.
If charged based on requests, Cursor considers it to be problematic; it can be completely changed to charge based on tokens, which is more transparent
I’ve found that Cursor’s prompt has changed since version 0.45. It’s something like “no rambling, just answer what’s needed.” This usually leads to cost savings on tokens (for Cursor) but gives users poorer results. Many tests show that limiting the model degrades its capabilities.
The tool calls in Claude-sonnet-max and Gemini models being charged at $0.5 each is too costly. Tool usage should also be counted within the 500 requests included in the normal Pro model. Sometimes, a single tool call results in a significant cost, which discourages users from using the max mode.
lol, happen to me, gemini not auto apply even in agent mode on new or existing chat
so I need to cost 1 request again just to ask gemini applying in last response
Would it be a solution to go to a previous version?
I have the same problem you guys mentioned above
My guess is that they’re experimenting with system prompts and other techniques to keep the context length as small as possible, while at the same time getting as much as possible from the models. Think about how compression works in terms of loss/gain. The reason for this is to keep down the costs of the API calls.
If you’ve ever tried for example Cline, you know how expensive it gets when you don’t care about managing the context. I used Cline for a specific problem today, and I spent $36 just in 30 minutes since I forgot I was not using cursor 
Yes, I can imagine that too.
But it affects our work so I hope that change to be done safely.
Is it guaranteed that such a change will not be introduced if we don’t do an upgrade? > Cursor team
I couldn’t agree more.
It reminds me a bit of the days of phones with a specific package like 100 sms and 2h of calls per month.
We’re just at the beginning and I hope we’ll just have a subscription one day with the unlimited package (an unrestricted and complete package)
Without disclosing the internals, to assist all that have an issue, add the following to your cursor rules:
<custom_instructions>
Immediately after any edit to a file, if the changes are incomplete, incorrect, or cause issues, please reapply the last edit to the same file to ensure the intended modifications are correctly and smartly implemented.
</custom_instructions>
Make sure your server are not using the files. I always stopped frontend and backend and since then, i don’t have this problem anymore. I use Cursor on a Windows 10 machine
I noticed this too. And it’s the main reason I’m using Cursor. I wonder how they even make a profit considering this? API costs seem so disproportionately higher than the subscription/fast request cost. This makes me think they are operating at a loss for a lot of their users so they can grow at this stage, but it also makes me fear for their future pricing model.
Yea, I believe they are loosing money. It’s just funny when people say things like “I’m paying $20/month for Cursor, and fast requests just disappear. Make it Cheaper!” 
Check this topic as we discussed this issue: How to make Cursor more aggressive
@maxfahl its not only to reduce context, forum was spammed by people complaining of “Verbosity” and they need to find a common balance, would I prefer they let us choose? sure, lets open a feature request and I’ll vote
Yes, this always happens to me, causing me not to like using Gemini Pro 2.5
Well I understand. But it is NOT Acceptable at all if this do such an experiment without any option to opt out.
We lose money & time on this bug.
Same same, almost not usable and extremly slow