( My Psychology Is Broken ) Request for Removal of Useless GPT-4.1 from AutoMode

Feedback
1

AutoMode is practically useless in large-scale projects, and the main reason is the inclusion of GPT-4.1—a weak and underperforming model—under the label of a premium feature. It completely halts our workflow.

I have to repeat the same instruction 20 times. It says “I’ll do it” and then ends the conversation. It’s ridiculous and extremely frustrating.

I sincerely hope the Cursor team has actually used this mode before suggesting it to us as a reliable solution.

I guess a regular word understanding mechanism I made could repeat itself back to me like a parrot, but I’m sure it would help me better than a model like GPT4.1.

2 Likes
2

It’s isn’t useless

3

GPT4.1 is a useless model. This is clear. This is 100% accurate information.

For some reason, when I give a task, Claude handles it perfectly in 600 lines, while GPT-4.1 throws back 20 lines — sometimes not even that. Maybe it’s because it’s just too amazing. I’m honestly starting to think Cursor included this model in Auto Mode just to push us toward switching to Ultra out of frustration.

2 Likes
4

No one in cursor team have b-al-ls to say yes it is only GPT 4. I just read all the topics and everyone answer around it without giving a real answer.

If it is not so why each time I choose a modale 3.5 or 3.7 it says: You’ve hit your usage limit
You’ve saved $83 on API model usage this month with Pro. Switch to Auto for more requests or set a Spend Limit to continue with Sonnet. Your usage limits will reset when your monthly cycle ends on 8/4/2025.

In the message saying “sonnet” without mentioning any modal “3.5 or 3.7 or whatever” thats is mean that auto mode is trash it is without sonnet and made me doubt that auto is cursor small “which is theirs” unless they prove that.

So what is your next move Cursor?

1 Like
5

I want to add something else after i made a test now on each modal that was included in auto mode and each time i select and generate it says:
You’ve hit your usage limit

You’ve saved $83 on API model usage this month with Pro. Switch to Auto for more requests or set a Spend Limit to continue with this model. Your usage limits will reset when your monthly cycle ends on 8/4/2025.

So stay only GPT 4 and cursor small “even deepseek is not working”. after my test it confirms my thoughts that auto now uses only GPT4 and cursor small. Cursor team if you dissagree with me so why it is not working?

1 Like
6

Alright folks, welcome to hour three of my thrilling war against AutoMode. I’ve now passed the 200-request mark just to get a simple, microscopic change done.

And what does this “state-of-the-art” model keep telling me?
“I’m on it, I’ll let you know.”
Translation: “I’ll waste more of your time and maybe, just maybe, vomit out 20 lines of half-assed code if I feel generous.”

Is this a joke? No—this is an insult to the concept of productivity. I kept trying out of some pathetic hope that it might actually do its job.
Big mistake.

AutoMode is an absolute disgrace. A useless, time-sucking black hole. Burn it, bury it, forget it ever existed.
Total trash.

Look, I’m sorry Cursor team—but please, for the sake of our sanity, stop recommending AutoMode. Either fix it properly, or just admit it’s broken beyond repair.

Because if I see that thing struggle one more time to handle the most basic request, I swear I’m going to have a full-blown meltdown and headbutt my monitor.

This is not a feature. This is psychological warfare.

My mental health is hanging by a thread here.

5 Likes
7

Hey everyone, just a quick update: I’ve officially hit 800 requests in AutoMode trying to get a simple task done—and still, nothing works. What an absolutely brilliant mode. Truly incredible.

It even comes with the amazing perk of unlimited requests… because you’ll need every single one of them to maybe get half a result.

It’s so “good” that Pro mode now feels like the old Free plan—except worse. And honestly? At this point, the Free version is better. At least there, you get the same failures without paying a cent.

Until Cursor fixes this mess, I’m canceling my subscription.
Call it what you want—I call it a reaction.

Because clearly, we’re being ignored. Maybe this will get someone’s attention.

8

Am I right in thinking that instead of using a microscope, you’re trying to examine a bacterium with a magnifying glass and lamenting that the magnifying glass isn’t good enough?

9

No. This is completely the wrong mode.

It can’t even do a simple piece of code from scratch. I say “do it” and it gives me 20 lines. Hilarious. Even with the most detailed prompt spelling out every tiny requirement, this is what happens.

You try 20 times to explain it in the best possible way, and it says “okay I’m doing it” — then just stops. Fantastic model, I guess?

Funny how when I switch to another model, it immediately spits out 600 lines without a problem.

It’s not just that. It doesn’t even “believe” what I’m telling it about the model. Hahaha. I say it’s wrong or incomplete, I even give it an example code file to show what’s missing — and it insists there’s nothing wrong.

It’s a complete meltdown. Auto mode is trash. I’m starting to think Cursor deliberately made Auto mode this bad just to push us into Ultra mode. Sorry, Cursor team.

10

:thinking:

I use Auto for very quick, simple edits or when I need to quickly learn something about the repository. He’s really dumb, but he can handle these two tasks.
There are smarter models for more complex tasks (I already wrote about this in my guide).

11

The real issue is that after Cursor’s new pricing policy, everyone’s credits are getting used up incredibly fast, and the Pro subscription is basically equivalent to the Free one now. In fact, like I said above, Free is actually better than Pro at this point.

This shouldn’t be too hard for Cursor to understand. If people are going crazy on the Cursor forum instead of working on their projects, that’s already pretty easy to see.

Once your credits run out, you’re forced to use Auto mode. And as I mentioned, you can’t even use it properly.

Not a single one out of 800 requests works? Ugh… I wouldn’t even bother using Cursor for a simple 3-line task. Most people here are using Cursor to actually build and launch their projects.

Maybe the whole model selection system should just be removed from Cursor entirely. Just fix Auto mode and let everyone use that, consistently, on a single solid model.

I don’t know. But until these things get fixed, I won’t be renewing my subscription.

12

Pro is for hobby coding. Pro+ already allows you to work, although optimizing your work with Cursor. Ultra for those who work a lot on projects of high complexity.

13

Previously, Pro users didn’t have these kinds of complaints — because there was no such division. The pricing policy made sense. Sure, you were sometimes on a slower mode, but at least your projects weren’t left unfinished. You didn’t waste an entire month for nothing.

And let’s be honest — the absurdity starts right at the pricing. Why would I pay $20 to Cursor? For $20, I can use Claude or Grok and already have access to a powerful code editor. The whole point of using Cursor was the variety and the ability to keep things going smoothly.

The current pricing model is completely illogical and ridiculous.

14

I fully understand those who feel misled by the word “unlimited” when buying a plan.
But complaining about not being able to pay $60/month for a tool that enables you to earn many times more — that’s just odd.

If another product can give you more for less money, then use it. But are there any other products that can do the SAME for less money?

Claude is not the best LLM on the market right now. Grok is located inside its Web UI (as far as I know). Gemini CLI is also not the same as Cursor IDE.

15

Of course this new pricing model seems illogical, let me explain why.

I used to be able to use Cursor for $20 without any problems, right? But now it’s practically unusable—even the free version is better. Could I just get Claude or Grok instead? Yes, I can.

Claude isn’t the best model, but it’s one of the best. At least it is on Cursor. Grok, Claude, and Gemini 2.5 Pro are currently among the top models. Let’s be honest: OpenAI models on Cursor just aren’t good.

Now let’s talk pricing. Can I take Grok or Claude and integrate them into something like Void? Yes, I can.

Also, consider that Cursor’s Ultra plan is $200 a month—that’s $2400 a year. Instead, I could just buy a $50 or $100 plan on Augment Code and get equal or even better performance.

What about the Pro plan? You can’t even make it halfway through the month before you’re out of credits. But on Augment Code, with $50 you get 600 requests—which is honestly hard to burn through with a model that actually works.

So here’s the question Cursor really has to answer: why should anyone upgrade to PRO+?

If that’s the case, just get rid of the Pro plan entirely. Make PRO+ the new PRO, and price everything purely per request, the way Augment Code does.

I mean, look—we mostly use Claude for complex projects. But why should I pay $60 for that? Another competitor already offers a better model with 600 requests for the same price.

It’s not just that. We’re also comparing Cursor to competitors like GitHub Copilot, which has $10 and $39 subscription tiers. Even the $10 plan is way better than Cursor’s current Pro tier—it doesn’t leave you stranded halfway through your work. And at $39, you can easily use top-tier models like Sonnet and Opus.

Why should I pay for a bunch of different models that I won’t even use?

16

‘GPT-4.1’ is the dumbest model — it can’t even properly refactor PHPDoc documentation for functions, even when given clear instructions on what to do and exactly where.
As it performs actions, it gets dumber and forgets what it’s supposed to do.

Yes, you’re right — sometimes it takes over 10 prompts just to get it to do something simple.
Offering ‘AUTO’ with this model is a joke — and so is the mode itself.

The Cursor team has decided to mistreat us in a harsh and twisted way.
The sooner you stop paying for this, the sooner they’ll be forced to fix things and choose a direction that benefits us, not just them.

1 Like
17

Cursor is dying, there’s insane feature inflation, I don’t understand why you’re defending them.
Not everyone likes being screwed over, but there are exceptions willing to pay for it.

1 Like
18

We already did the detailed, mathematical breakdown above. Cursor is completely in the wrong on this.

19

On the old pricing model, I was paying around $30–40 a month. Now I’m paying $60, even though Cursor itself has improved and there are smarter models available.

If I were smart enough and didn’t waste my time on my open-source projects, I could probably be making at least $60 a day — so as soon as I stop messing around, I’ll most likely be able to afford switching to Ultra and finally stop refreshing the usage page.

You have to understand that the prices for using the IDE are based on prices from providers, and not drawn out of thin air. I’m saying this because I’ve been using another service with a similar business model for two years now, and they also provide access to neural networks cheaper than on public pricing pages.

1 Like
20

Not everyone needs Pro+. There are those who have enough Pro. There are those who want a cheap Tab plan at all.

2 Likes