Millions tokens without cache read and extremely high cost

[Chris_Zhu]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Cache Read 0
Cache Write 0
Input 3,134,794
Output 14,216
Total 3,149,010
Cost $16.03

Steps to Reproduce

None

Expected Behavior

Cache Read 2,744,297
Cache Write 112,200
Input 520
Output 27,651
Total 2,884,668
Cost $2.77

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.3.34 (Universal)
VSCode Version: 1.105.1
Commit: 643ba67cd252e2888e296dd0cf34a0c5d7625b90
Date: 2026-01-10T21:17:10.428Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Darwin arm64 24.6.0

For AI issues: add Request ID with privacy disabled

a124aa9e-e426-4d2a-af95-948a5db727ee

Does this stop you from using Cursor

No - Cursor works, but with this issue

[Artemonim]

You didn’t provide any steps to reproduce or the model you used.

[Chris_Zhu]

It just happened.

I just created a new chat session in agent mode with 4.5 opus, and asked one question(349k no cache) and a follow-up question(3.1M no cache).

Before this, the cursor cache read worked as expected.

[deanrie]

Hey, it looks like you have Privacy Mode enabled, which blocks us from diagnosing the issue on our side.

For the team to investigate, please disable Privacy Mode:

• Cursor Settings > General
• Privacy section
• Turn off Privacy Mode
• Reproduce the issue again
• Copy the new Request ID

We suspect summarization might be breaking caching for requests this large (3,1M tokens). That’s a really huge amount. Summarization usually kicks in for long chats and can change the context in a way that stops Anthropic’s cache from working.

You can also try:

• Starting a new chat instead of continuing a long one
• Using shorter requests

Send the new Request ID after turning off Privacy Mode, and the team can check the details.