Plan mode appears to fall into an infinite loop

[vvky]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Plan mode appears to fall into an infinite loop, resulting in massive token consumption. Repeated execution generates a large amount of accumulated context. In a simple task test, using Opus consumed over 3 million tokens.The task has already been completed, but it continues to be repeatedly checked.

Steps to Reproduce

When starting a task in plan mode, it may occasionally and inexplicably fall into an infinite loop. I have encountered the same issue on other accounts as well. Could this be related to my use of Chinese?

Screenshots / Screen Recordings

Operating System

MacOS

Version Information

Version: 2.4.21
VSCode Version: 1.105.1
Commit: dc8361355d709f306d5159635a677a571b277bc0
Date: 2026-01-22T16:57:59.675Z (3 days ago)
Build Type: Stable
Release Track: Default
Electron: 39.2.7
Chromium: 142.0.7444.235
Node.js: 22.21.1
V8: 14.2.231.21-electron.0
OS: Darwin arm64 24.6.0

For AI issues: which model did you use?

opus 4.5 thinking

For AI issues: add Request ID with privacy disabled

request id: b1cbabbf-4155-4680-87e1-0095b82da73d

Does this stop you from using Cursor

Yes - Cursor is unusable

[deanrie]

Hey, thanks for the report. I see the screenshot with the looping “thought” block, this is a known issue.

The team is aware: Endless thinking loop in Plan Mode subagent

Root cause is a bug in how Opus parses thinking tags. The model can’t exit the <think> block and gets stuck re-checking tasks that are already done. That explains the 3M+ token usage.

Workaround for now: Don’t use Plan mode with Opus 4.5 thinking for the moment. Or try another model (GPT, Gemini). Similar loop bugs can happen there too, but they’re usually not as bad with tokens.

Your request ID will help the team track the specific case.

About Chinese, it’s most likely not related. Other users hit the same thing in English too.