Cursor needs better pricing guardrails when one session can cost $600+

Ideas Feedback
,
1

I’m already on the Ultra plan and spend around $2,000/month on Cursor, so this is not a complaint about paying for value.

The problem is that pricing has become too opaque to use responsibly when a single normal-looking session can turn into a $600+ bill.

The main issues for me are:

  • Pricing visibility is too weak overall. With models like GPT-5.4-xhigh, cost is often not proportional to the visible amount of work getting done. Some models can spend a very long time thinking and make an insane number of tool calls before taking action. So you can’t reliably tell from the task whether something will cost a few dollars or hundreds, and there’s no easy way to spot runaway behavior early enough.
  • Cloud Agent forces Max mode. That used to make more sense than it does now. In my experience, it often doesn’t help much anymore, and can even hurt. Models often do better with summarized context than when pushed deep into very long-context behavior, where they get erratic. Around 200k context is enough for nearly all real work.
  • Cloud Agent is too rigid about model choice. Locally, I switch models all the time: GPT-5.4 or Opus 4.6 for planning, Composer 2 for implementation, then back to GPT/Opus for PR descriptions and polished writing. That workflow helps control cost without giving up quality. In Cloud Agent, that flexibility is missing.

I’m fine with expensive models existing. I’m not fine with a UX where:

  • cost is hard to predict
  • some models can silently burn money by overthinking and over-calling tools
  • Max mode is forced in Cloud Agent
  • and users can’t adapt model choice mid-run the way many of us already do locally

If Cursor wants usage-based pricing at this level, it needs better guardrails:

  • better pricing visibility while work is happening
  • better per-session cost estimates
  • visibility into runaway thinking/tool-call behavior
  • warnings for unusually expensive sessions
  • the ability to switch models mid-session in Cloud Agent
  • the option to run Cloud Agent without forced Max mode
2 Likes
3

Hey, thanks for the detailed feedback. It’s clearly written, and this kind of specific feedback really helps.

A couple things to know:

Existing tools: if you haven’t set this up yet, your dashboard has spend limits and spend alerts.

These work at the monthly level, not per session, so they won’t fully solve what you’re describing, but they can at least limit overall risk.

On your specific requests:

  • Per-session cost visibility, “runaway” spend warnings, and clearer pricing during a session are on our radar.
  • Switching models mid-session in Cloud Agent: agreed, this would be useful, especially since different models fit different stages of work. The team is aware of this gap.
  • Forcing Max mode in Cloud Agent: fair point, especially given how context handling has evolved. Noted.

I’ll move this to Feature Requests since you laid out clear, concrete requests that users can vote on.

Let me know if there’s anything else.

4
5

I don’t understand what the savings that this cursor tokens website shows are supposed to be.

6

The more data you have, the better you make decisions about optimizing usage. This site provides much more data than the official dashboard.

7

Wow this is serious amount of money!!! 600 for Cursor alone? Wow.

I’ve been a heavy user since August. And at the way the new Composer2 is making serious lapses, this is seriously a lot of waste… You need to look at other AI tools out there and compare before you spend more on Cursor. That’s really a lot for the quality of work you are getting.