Cursor using MAX even when disabled

[wpozzatti (Wendell Mosquini Pozzatti)]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Hi everyone,

Since last month, I’ve noticed that my monthly Cursor limits are being consumed abnormally quickly. In November, I upgraded to the Pro+ plan, expecting an improvement, but the consumption has become even faster.

My December data allowance was renewed on December 14th, and in less than three days, I started receiving usage alerts.

Cursor recently released the Claude Opus 4.5 for use without MAX, but this model is consuming the limits far above normal. I emphasize that I have never used MAX on my account. There is no record of this in my usage history. I only enabled the Opus model, just as I do with Sonnet, without activating the MAX feature.

I contacted support by email, but the response has been just a back-and-forth of messages without a concrete solution, which has been quite frustrating.

I suspect there’s a bug in the Cursor and that Opus, even with MAX disabled, is consuming resources as if it were active.

I’m considering canceling, but I’d like to know if anyone else is experiencing this same problem.

Steps to Reproduce

Normal AI usage. I just went into the settings, enabled the Opus model, and then selected it in the chat.

It’s important to mention that Opus used to show the MAX model for use. However, it doesn’t show it now, which suggests it’s set to a conventional model, like the Sonnet, GPT, etc.

Operating System

Windows 10/11

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.2.43 (user setup)
VSCode Version: 1.105.1
Commit: 32cfbe848b35d9eb320980195985450f244b3030
Date: 2025-12-19T06:06:44.644Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Windows_NT x64 10.0.26200

For AI issues: which model did you use?

Opus 4.5

Additional Information

Using models WITHOUT MAX activated

Does this stop you from using Cursor

No - Cursor works, but with this issue

[deanrie]

Hey, thanks for the report.

Opus 4.5 can burn through usage faster than other models, but that isn’t a bug. Usage is calculated from tokens at the model’s API rates, not from a fixed number of “requests”. Opus is more expensive on the API, so limits get used up faster.

A few important things to check:

1. Make sure MAX Mode wasn’t enabled by accident. The MAX Mode button is very close to the Opus model picker, so it’s easy to click it by mistake. Other users mentioned this here:
   Cursor’s Undisclosed Model Upgrade Issue

2. Check the detailed breakdown. Go to https://cursor.com/dashboard?tab=usage and review the per-request token breakdown (input/output/cache). That should also make it clear whether MAX Mode was used.

3. How the counting works. I explained it in more detail here:
   Opus 4.5 Doesn't Show Requests Count as 2x Requests - #5 by deanrie

4. Official docs:

• Models and pricing: Models | Cursor Docs
• Plans: Pricing | Cursor Docs

If, after checking the dashboard, you’re sure MAX Mode wasn’t used but usage still looks abnormally high, please share:

• A screenshot of the usage breakdown from the dashboard
• Request IDs for a few affected requests (in chat: top-right menu → Copy Request ID)
• Rough context size, like how many files were attached and the approximate repo size

Let me know what the dashboard shows.

[Serp]

I can see same thing happening. Randomly usage being marked as MAX, but not using that setting.

[wpozzatti (Wendell Mosquini Pozzatti)]

@deanrie Thank you for your reply.

I am aware that Opus has higher resource consumption, but the main problem is that the consumption appears to be based on MAX. However, as I mentioned, I don’t use MAX and the option is completely disabled.

The only procedure I perform is switching between models.

Like @Serp , MAX is disabled; the history on the Cursor dashboard confirms this.

I’ve attached my dashboard below. The records with MAX were only for testing purposes to show this to Cursor’s email support.

Strangely, I suspect this is a Cursor problem. Previously, Opus only worked with MAX enabled, but apparently the model was released without dependencies, but with very high resource consumption.

Interestingly, notice that the last two records in the image are from the same quick test, where the AI ​​wrote practically the same thing. One has the MAX value and the other doesn’t, yet they consumed the same number of tokens.

[erraoudy (Said)]

Hi @deanrie

I’m experiencing the same issue, Cursor keeps automatically switching my default model to Claude-4.5-Opus-High-Thinking, even though this model is disabled in my settings (see screenshots attached).

Every time I switch from Agent to Plan, Cursor silently resets this model as the default, which results in hundreds of unintended, expensive requests. I did not authorize these calls, and I’ve lost a significant number of plan requests because of this hidden behavior.

Frankly, this doesn’t make sense. As a paying user, I should have full control over which models are active. Being forced to use a disabled model feels deceptive and unacceptable.

Please investigate this and ensure it’s fixed as soon as possible.

[DemonVN]

Agents often try to give long and detailed answers, which significantly consumes your tokens. Follow my post and apply this rule to your cursor; it will definitely save you a considerable amount of tokens.

I can use Opus 4.5/Sonet 4.5 thinking at a cheaper price

Hi everyone, We will have to pay for both input and output. Input is usually cheaper than output, but input is cached while output is not. This means you will pay 100% token fees for output. So, if we instruct the Agent to minimize output, we will save a lot of tokens. The rule I created aims to minimize the Agent’s output while maintaining maximum reasoning/logic/code/intelligence, and avoiding wasteful token behaviors such as summarizing tasks after completion, creating unnecessary .md and .t…

[Vasisht_Sankar]

I have the same issue. When trying to use opus 4.5 all requests are opted into MAX mode without my consent. I had to go to https://cursor.com/dashboard?tab=usage to figure out what was happening, only to see my tokens burning abnormally high.

At least add an indication on the cursor agent cli that this is happening so folks are not misled. And please fix the behavior for non ide users like us, this is really frustrating.

[visited-redirect]

I have the same issue. I never turn on max mode but I got randomly charged for it. It burned through 61.8 requests in 5 chats!

This is a serious problem and we should get refunded for that.