Can't do basic math

Support Bug Reports
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Prompt: 1 tick is 1 microsecond. It will advance the timestamp by 1000.
Output:

Confirming: 1 tick = 1 microsecond = 1000 nanoseconds So:
19900->20900: 10000 ns = 10 ticks
20900->21900: 10000 ns = 10 ticks
21900->22900: 10000 ns = 10 ticks
The fix is correct. Running the test:

Steps to Reproduce

I was in the middle of a complicated debugging session. I am not sure if a simpler context would reproduce or not. Its ridiculous. I was using 1x agent on Composer-1 model.

Expected Behavior

20900 - 19900 = 1000.

Screenshots / Screen Recordings

CursorMathBug.png
CursorMathBug.png396×192 23.4 KB

Operating System

Windows 10/11

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.3.15 (user setup)
VSCode Version: 1.105.1
Commit: bb2dbaacf30bb7eb9fd48a37812a8f326defa530
Date: 2025-12-30T20:30:37.151Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Windows_NT x64 10.0.26100

For AI issues: which model did you use?

Composer 1

For AI issues: add Request ID with privacy disabled

Unknown and cant disable privacy mode without permission.

Additional Information

I don’t feel comfortable disabling privacy mode without permission from my company. I may have also reported the issue with command palette and report AI issue bad action.

Does this stop you from using Cursor

No - Cursor works, but with this issue

3

LLMs cant do math. not even the simple math. some SOTA models (gemini 3 pro, opus 4.5) are a bit better at it though. yet I would not trust it do any math
as far as I am aware composer model is strictly focused on coding tasks so not much of a surprise

4

Hey, thanks for the report. This is a known limitation of LLM models. Even the most advanced models can make arithmetic mistakes, especially in the middle of complex debugging sessions.

A few suggestions:

  • Try stronger models (Claude Opus 4.5 or Gemini 3 Pro) for tasks that need accurate calculations.
  • Ask the model to show the math step by step: “Show your calculation step by step”
  • For critical math, it’s best to double-check manually or use separate tools.

Composer-1 is optimized for coding tasks, so arithmetic errors during debugging aren’t really a bug, they’re more a general limitation of current LLM models.