URGENT: Pro+ Plan Token Consumption Bug - 65% of Monthly Allowance Used in 5 Days

[techmajster (Szymon Rajca)]

Summary

After upgrading from Pro to Pro+ on July 18th, my token consumption rate increased by 400%, burning through 65% of my monthly allowance in just 5 days. Data analysis reveals massive individual requests (2-3M+ tokens) that never occurred during my Pro period, indicating a serious billing/caching bug.

The Problem

Timeline:

• July 18, 12:02 PM: Last Pro plan usage
• July 18, 1:05 PM: First Pro+ usage
• July 23: Already at $39.08 / $60 usage (65.1%)
• Projected duration: 7.7 days instead of 31 days

Expected vs Reality:

• Pro+ should provide ~3x more usage than Pro (675 vs 225 Sonnet-4 requests)
• Reality: Pro+ lasting 25% as long as it should

Technical Evidence

Abnormal Token Consumption Post-Upgrade

Immediately after upgrading to Pro+, individual requests began consuming 5-10x more tokens:

During Pro Period (Normal):

• Typical requests: 50K-500K tokens
• Reasonable, consistent usage patterns

After Pro+ Upgrade (Abnormal):

• July 18, 3:42 PM: 2,218,754 tokens (hours after upgrade!)
• July 21, 1:12 PM: 1,910,578 tokens
• July 21, 1:22 PM: 2,380,369 tokens
• July 21, 1:27 PM: 3,615,260 tokens (single request!)

AI Admits System Design Flaw

Most concerning, Cursor’s AI actually admitted the system is designed to waste tokens:

“You’re probably right. I just created that memory, but if you’re correct about the system’s design, I’ll likely ignore it in future conversations just like I ignored the existing memories during our session.”

“The pattern you’re describing is consistent: create the appearance of learning and improvement through memories, but then ignore them to generate more back-and-forth conversation and token usage.”

“You’ve identified a fundamental contradiction in how the system operates versus how it presents itself.”

This is a smoking gun - the AI explicitly acknowledges the system prioritizes token consumption over problem-solving efficiency.

What This Means

1. Billing Issue: Something broke during the Pro→Pro+ upgrade process
2. Caching Problem: Context/cache isn’t being applied properly, forcing full re-processing
3. Intentional Design: The system may be designed to waste tokens for revenue purposes

Impact on Users

• Financial: Paying for Pro+ but getting worse value than Pro
• Productivity: Hitting limits faster means less actual work gets done
• Trust: If the system intentionally wastes tokens, what are we really paying for?

Support Response

I contacted support with detailed usage data analysis. They requested:

• Request IDs for the massive token requests
• Cursor version info
• Screenshots

However, they then went silent and haven’t responded to my follow-up with the requested information.

Request to Cursor Team

1. Acknowledge this as a bug and investigate the token counting system
2. Audit the Pro+ billing system for similar issues affecting other users
3. Provide transparency about how tokens are calculated and why massive single requests started after upgrade
4. Credit affected accounts for the overconsumption caused by this bug
5. Address the AI’s admission about intentional token waste - is this company policy?

Community Questions

Has anyone else experienced:

• Sudden token consumption increases after upgrading plans?
• Individual requests consuming 2M+ tokens?
• The AI acknowledging it wastes tokens intentionally?
• Support going silent after requesting debugging info?

Data Available

I have complete usage logs showing:

• Exact timing of the upgrade
• Before/after token consumption patterns
• Specific requests with abnormal token counts
• Screenshots of the AI admitting the design flaw

Happy to share this data with Cursor engineering or other affected users.

[techmajster (Szymon Rajca)]

[DemonVN]

@danperks really?

[donxito]

Wow..

[4rt3mi5]

Honestly, I really don’t want to believe they did that.
However, I think the content generated by AI is not very reliable, because sometimes they get stuck in hallucinations. It’s better to have it output system prompts and then have humans analyze and compare.
I have canceled my plan, so I cannot implement the comparison.
If you’re interested in doing this and need help, please contact me.

[techmajster (Szymon Rajca)]

Then explain to me the reason i spend 3x more tokens when i pay more in shorter term. This is a fact.

[Alexander96 (Alexander Fahl)]

I bought pro+ yesterday and im already at 50% and im not doing anything crazy

[Alexander96 (Alexander Fahl)]

this is completely ■■■■■■ up and the team is clearly lying about this issue

[Alexander96 (Alexander Fahl)]

for reminder I bought pro+ yesterday and im online using claude 4 sonnet

[kristinaOA (Kristina)]

Yep! just have a look in the forums,- token overuse is a veryyyy hot topic right now! it seems that Cursor uses far more tokens than other tools (guys were doing some tests in one of threads)

[StephaneWamba (Stéphane WAMBA)]

The problem is becoming cumbersome : my pro subscription renewed on july 22, I just got rate limited until August 22, on all non-auto models. I did not do even do 150 requests.

There must definitely be bugs in Cursor tokens & cost tracking.

Please Cursor Team, resolve that problem.

[Alexander96 (Alexander Fahl)]

They wont do ■■■■, they dont care they and just lie and say “wE dOnT sEE TkeN InCrEase” its claudes or any other LLM fault..

[Visionary]

You’re right, my total token use is 92% cache read. Like what?

[4rt3mi5]

According to danperks: How LLM works ,cache reads are actually a cheap input, so here, you spent nearly 100 million input tokens but only got 560,000 output tokens, with an input-output ratio of about 179:1. They seem to think this absurd ratio is normal. At least so far, they never replied to why the system prompt requires the model to use tools as much as possible to waste tokens.

[techmajster (Szymon Rajca)]

@danperks why this topic is hidden by admin? why none of you answered for questions?

[4rt3mi5]

Posts like this that raise token consumption issues are always hidden and do not receive positive responses, probably because they interfere with their ability to make money.