The Cursor is amazing… But the plans make it unusable

my experience, Cursor’s $20 or $60 plans are practically unusable. You subscribe to something that’s supposedly monthly, but the limit runs out in just a few hours or, at best, a few days, depending on your usage. After that, you’re basically stuck.

The only alternative is the automatic mode, which isn’t automatic at all, since it only uses the Composer model — and honestly, it’s extremely weak for more complex tasks, especially backend-related ones. It can handle simple and quick things, but for anything more serious, it’s useless.

Because of this, I abandoned Cursor and switched to Antigrafic. At least there, the limits are more realistic, and when you hit them, they recharge within a few hours, which makes a lot more sense for a monthly plan.

I’ve been using Cursor since the beginning: their billing model is terrible and ends up making the tool practically unusable.

1 Like

Hey, thanks for the feedback. I get that limits can run out fast depending on your usage patterns.

A few notes about the limits system:

  • The burn rate depends on the models you pick. Top models (Opus, GPT-5) use limits faster than lighter ones. The dashboard shows a detailed token breakdown: https://cursor.com/dashboard?tab=usage
  • When your included limit runs out, you can add on-demand usage and keep going at the same rates (pay-as-you-go).
  • From our stats, Daily Agent users usually spend $60 to $100 per month in total usage, and Power users $200+.

About Auto mode, it does use balanced models to optimize cost and performance. For harder backend work, you can manually pick a top model (Opus 4.5) in chat.

I’ll pass this feedback to the team. We track all pricing comments, but I can’t share any timeline for changes.

@deanrie I just signed up for Cursor. I’ve been using it for just a little under 2 hours. Let’s just call it 2 hours. In Agent mode with the Cursor “Composer 1” model and I have already racked up $3.97. Nothing too difficult. A code base that’s only 2 weeks old, one developer, maybe 20 files. During these 2 hours I’ve been working with the Agent on a Python Textual TUI, rendering data to screens and creating some tests. $4 over two hours is cheaper that a dev salary, but quite a bit more than Claude Code. I cannot justify this cost. I will not renew after this first month.

Hey, thanks for the feedback.

$4 for 2 hours is definitely fast for a relatively small codebase, I get how that can feel expensive. Agent mode with the top models burns through limits faster than regular chat because each iteration includes more context and reasoning.

A few things that can help lower costs if you decide to keep going:

  • Try Auto mode instead of using Composer 1 manually, it picks lighter models for simple tasks
  • For simpler tasks, use Chat with cheaper models (Flash, Mini)
  • The Dashboard shows a detailed token breakdown, so you can see exactly what’s using tokens: https://cursor.com/dashboard?tab=usage

I’ll pass your feedback to the team.