This happens equally with all plans. Mine are with Claude Sonnet 4.5. Also, if you have the plan window open while Cursor is writing, you can see it attempts to redraw the page with every keystroke. Perhaps this is the source of the problem. Opening a different window removes the visual flickering effect, but the plan writing is still so slow as to be nearly useless.
Plan mode 5-10 minutes streaming the plan file, Ask mode 10 seconds. So the performance issue of plan mode is HUGE. I did not include Thinking as GPT 5.2 High, thinks a lot of time, that is fine.
80K context window during plan generation, total size of codebase reachable by the Agent is about 10000+ files because i have several submodules and cloned repos i use to allow the agent to see the implementation of stuff I use (eg vscode codebase for monaco) and their docs (eg aisdk and tiptap docs). My code is about 100 files though. However i want to emphasize that i do not have a performance issue with Cursor in general, everything is very fast for me, its ONLY Plan mode that in the past 2 weeks became a snail and have several issues, slowness is only the worse one, like is so bad that it made me open this report.
On top of it while plan mode is generating the file, the chat panel is unusable, everything keeps re-rendering and i cannot keep the context window open to copy the request id, i had to wait for it to finish or stop it.
In general I would say that the quality of Plan mode whent down drastically in the past 2-3 weeks, I started to have problems I did not have before making it quite frustrating to use, and I’ve been using it from day one, when it was in Early Access, it gave me much less headache back then.
This test was done with GPT 5.2 High Fast
I also attach 2 videos that clearly shows the difference.
Thanks for the detailed info and the video. The Request ID and timing comparison will really help the team.
I can see the issue is confirmed by other users too, including visual flickering while generating the plan. I’ve passed all the details to the team to investigate. 5 to 10 minutes vs 10 seconds is a critical difference.
I also noted that the re-rendering issue makes the chat panel unusable while the plan is being generated.
We’ll share updates as soon as the team has more info.
