lagging system
Open 💬 4 comments Opened Aug 14, 2026 by nukfulkchkbuk
💡 Likely answer: A maintainer (github-actions[bot], contributor)
responded on this thread — see the highlighted reply below.
What version of the Codex App are you using (From “About Codex” dialog)?
Version 26.810.41047Version 26.810.41047Version 26.810.41047
What subscription do you have?
Plus
What platform is your computer?
Win 11 Home 64 bit
What issue are you seeing?
Since a recent update, after processing a task. The mouse gets super laggy and jumps around the screen. System is fine after closing Codex. The bug is not too far from rendering my PC unusable.
What steps can reproduce the bug?
On startup and after working on a task.
What is the expected behavior?
No lag
Additional information
Not sure about this part so take it with a pinch of salt: I think my NVME drive is being hammered IO wise not bandwidth when Codex is idle.
4 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
same issue here,
I’m switching to Claude Max.
What the hell are the OpenAI devs doing?????
I've been seeing the same issue. I have submitted it with feedback ID no-active-thread-01a004fc-8796-7711-b0ae-c6f6973ff875; I attach the report also as a markdown file. Relevant screenshots and a screen capture of the lag vanishing when switching to Task Manager I have included via https://imgur.com/a/J3QXIc1.
no-active-thread-01a004fc-8796-7711-b0ae-c6f6973ff875.md
I can reproduce this on Windows with an RTX 5070 Laptop GPU, and I was able to narrow the trigger down further.
Observed behavior:
This strongly suggests that, at least on my machine, the trigger is the visible Codex workspace rendering/compositing path, rather than an active agent task, conversation length, or general CPU saturation. The fact that mere window occlusion eliminates the stutter seems especially diagnostic because Codex remains open and running.
This looks closely related to #38710 (GPU/DWM load), but the occlusion test may help isolate the rendering/compositor side of the regression.