Rate limits spends after few requests

[wtester]

I think it’s pointless to try to prove anything — the issue is being denied, and we’re told that everything is working as intended.
We’re paying for input and output, but on top of that, we’re given a nice little bonus in the form of some cache, which multiplies the cost of a request many times over.

[thepKz (Thep Mai Tan)]

@condor

Cache Read is part of chat session that is already cached at API provider, this avoids repeated processing as it has been ingested for the session. provider reads from cache and uses it as context for response. Cache Read is cheapest of all the 4 columns.

I do not think this is from previous sessions, could you confirm if there is any relation of new project to old projects and if there is any other project in Cursor which might have used similar code or still has files indexed?

I created a completely new project, and the codebase indexing is 0 — meaning the folder is entirely empty.
I noticed that the token usage is calculated as the sum across all 4 columns, as you can see in the image. After switching to default and gpt-4.1, the Cache Read value became almost negligible.

I’m quite shocked that for three consecutive days, I only made around 15 requests using claude-4-thinking, yet things have gotten progressively worse. I keep hitting rate limits, and I really hope this can be clarified as soon as possible.

[condor (Condor (Tee))]

Cache

• Ok it would not have been necessarily codebase indexing of the new project, but I dont have insight into which steps were run in your chat.
• When switching providers (claude vs gpt) there can not be any cache transfer
• I have seen some issues with claude too today, not sure how much of that is Anthropics part, but its not likey Cursor 1.2 as i have been testing it with same models already before.

Tokens

• Yes token usage is a sum, the price would match the pricing on this page +20% per docs Pricing \ Anthropic for Claude models.

Usage

• As for usage, I do not have myself any jump in token usage but given a few reports it may need to be checked more. Im gonna inquire but may be a bit time until I get an answer. Its holiday and I’m also gonna be AFK.

[FlamesONE (Nikita)]

2 messages and again.

[nikjankov]

I’ve been locked out of Cursor’s Claude 4 Sonnet for 29 hours and 2 minutes now, which suggests Cursor’s rate limiting implementation is significantly different from Anthropic’s actual system.

When using Claude directly through Anthropic’s interface, the rate limiting behavior is much more transparent and reasonable:

• I receive clear notifications when approaching limits (e.g., “You’re near your rate limit at 2 PM, it will refresh at 3 PM”)
• Rate limit resets typically occur within hours, not days
• The system provides specific timeframes for when limits will refresh
• Token consumption appears more consistent and predictable

The fact that I can have normal conversations with Claude directly while being completely locked out of Cursor for over a day indicates that either:

1. Cursor is implementing additional rate limiting on top of Anthropic’s limits
2. There’s a bug in how Cursor is calculating or applying these limits
3. Cursor’s context management is inflating token usage beyond what Anthropic actually charges

For comparison, Anthropic’s published rate limits for Claude models typically reset within 24 hours maximum, and users get clear visibility into their usage and reset times. A 29+ hour lockout with no clear reset timeframe doesn’t align with Anthropic’s documented behavior.

The Cursor team should clarify whether they’re adding their own rate limiting layer or if there’s a technical issue causing these extended lockouts that far exceed Anthropic’s actual limits.

[condor (Condor (Tee))]

further updates on this topic

Clarifying Cursor Pro Pricing

We want to acknowledge that our recent pricing changes for individual plans were not communicated clearly. That’s our mistake, and we’re sorry. If you were unexpectedly charged for usage between June 16 and July 4, we’ll issue a full refund - just email us at [email protected]. Here’s what’s important to know about the Pro plan: Unlimited usage of Tab and models in Auto $20 of frontier model usage per month at API pricing An option to purchase more frontier model usage at cost We’ve br…

[FlamesONE (Nikita)]

Yea, I saw.
Old pro plan / 2 = new plan.

Cursor was a great product. Will wait for changes, but for now, I’m going to search alternatives