Codex slowness - "Thinking" state takes very long while the actual actions are fast

What version of the Codex App are you using (From “About Codex” dialog)?

Version 26.623.101652 • Released 3 juil. 2026

What subscription do you have?

ChatGPT Plus

What platform is your computer?

Microsoft Windows NT 10.0.19045.0 x64

What issue are you seeing?

ID of issue : 019f3691-4331-7db2-a41f-7289d5776707

Since yesterday, I use Codex to assist me in correcting my students' exams.
I live in Congo (DRC) and is under the UTC+1 timezone. Yesterday morning each exam corrected by Codex lasted around 1 minute 30 sec. 2 hours later, the time went up to almost 10 minutes per exam. 2 hours later, each correction took almost an hour.
I assumed that it was because the USA was waking up as for me it was the evening and for them the morning.

This morning the scenario is almost the same except that instead of 1 minute in the morning, the tasks lasts around 3 minutes and 30 minutes later, it went up to 15 minutes per tasks.

As symptom: I noticed that Codex performs actions quickly once it executes them, but remains abnormally stuck between actions on "Thinking" ("Réflexion en cours" in french) for an unusually long time. This seems more like orchestration, capacity, compaction, or account limitation latency than actual model reflection. As explained, the issue affects repetitive tasks that were previously fast, and similar reports exist on Discord as well as in the OpenAI/GitHub public channels.

I mainly use GPT-5.5 in standard and medium mode. I tried switching to 5.4 or 5.5 low but the waiting time stays the same (no improvement).

What steps can reproduce the bug?

Any tasks takes very long to accomplish.

What is the expected behavior?

_No response_

Additional information

_No response_