Athropic modles using millions of tokens and zero cache reads in Cursor

[CosmoBro]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Anthropic models have gone off the rails and are using zero cache context and pulling MILLIONS of tokens in input! WTF is going on here!? I have been using these models nonstop for the last year with zero issues. The only change I’ve made is making use of MCP tools for viewing images.

I will need to leave cursor if this isn’t possible to solve because the Anthropic models work better for me than any other option by far.

Steps to Reproduce

Select sonnet, write any prompt and boom, millions of input tokens.

Screenshots / Screen Recordings

Operating System

Windows 10/11

Current Cursor Version (Menu → About Cursor → Copy)

Version: 1.4.2 (user setup)
VSCode Version: 1.99.3
Commit: d01860bc5f5a36b62f8a77cd42578126270db340
Date: 2025-08-07T17:16:23.005Z
Electron: 34.5.1
Chromium: 132.0.6834.210
Node.js: 20.19.0
V8: 13.2.152.41-electron.0
OS: Windows_NT x64 10.0.26100

For AI issues: which model did you use?

Sonnet 4

Additional Information

I will need to leave cursor if this isn’t possible to solve because the Anthropic models work better for me than any other option by far.

Does this stop you from using Cursor

Yes - Cursor is unusable

[sanjeed5 (Sanjeed)]

Thanks for reporting the issue. Can you check the token usage without enabling the MCP tools? You should be able to see the context window here.

[CosmoBro]

I didn’t do anything and now it is behaving normally again with MCP tools enabled and disabled… This ate up my entire ultra subscription budget in a matter of a few days because I didn’t realize what was going on, and my usage here (screenshot)

looked very flat. Can you guys help me out with a credit since their was an issue with the platform? I really don’t understand what could have caused this behavior.

[CosmoBro]

Intentsive prompts are now behaving the way the should