5.3-codex(xhigh/high/medium) has become slow + does not update progress as often

Resolved 💬 15 comments Opened Feb 9, 2026 by kiranjd Closed Feb 14, 2026

💡 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)?

260208.1016 (571)

What subscription do you have?

Pro

What issue are you seeing?

Since today morning(11am IST), 5.3 model has been acting very weird. It is more like codex-5.2 - slow and concise.

I have been using 5.3 since release and the model I am using sure is not the same as it was last 2 days. I have tried:

updating to latest verions
Changing models, thinking
Logging out and logging in

Thread id: 019c4127-c856-7853-935b-a30879955ae7
<img width="752" height="505" alt="Image" src="https://github.com/user-attachments/assets/8b845a52-1910-4072-a4e9-106a751113ea" />

What steps can reproduce the bug?

Just run any task

What is the expected behavior?

_No response_

Additional information

_No response_

View original on GitHub ↗

15 Comments

github-actions[bot] contributor · 5 months ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

#11189
#11187

Powered by Codex Action

theBucky · 5 months ago

You should probably trace the SSE events via codex cli.

I had this as well, and it turns out that the ChatGPT backend routed my request to gpt-5.2, despite me choosing 5.3-codex.

kiranjd · 5 months ago

yes, that was it

Your ~/.codex/config.toml is set to model = "gpt-5.3-codex", but when I traced the actual SSE “response.created” event, the backend reported it
  was using:

  - response.model: gpt-5.2-2025-12-11 (not gpt-5.3-codex)

  That matches the duplicate reports (openai/codex#11189, #11187) and would explain “feels like 5.2”.

  To capture debuggable proof (model + response id) without dumping your whole prompt:

  RUST_LOG='codex_api::sse::responses=trace' \
    codex exec --skip-git-repo-check -m gpt-5.3-codex "Reply with exactly: PING" 2>&1 \
  | python3 -c 'import json,re,sys
  for line in sys.stdin:
    m=re.search(r"SSE event: (\{.*\})", line)
    if not m: continue
    evt=json.loads(m.group(1))
    if evt.get("type")=="response.created":
      r=evt["response"]
      print("response.model:", r.get("model"))
      print("response.id:", r.get("id"))
      break'

  Also include in the GitHub issue:

  - codex --version
  - the printed response.model + response.id
  - your existing thread id (already in openai/codex#11215)

qwang961-a11y · 5 months ago

I have this as well. Pro subscription. very weird

zenabot27 · 5 months ago

Same here

jskorlol · 5 months ago

The performance has noticeably worsened compared to before — it’s slower, less accurate, and feels worse than it did back when Codex 5.2 was new.

LoaderTime · 5 months ago

Same. It keeps looping in searching and unable to update progress.

yowave · 5 months ago

Same, for me gpt-5.3-codex routes to gpt-5.2-2025-12-11.

Xayaan · 5 months ago

Same here gpt-5.3-codex

bkrescaperaseuse · 5 months ago

Go like/bookmark this reply from Tibo X post please: https://x.com/kappax/status/2021224883326173319

MisterRound · 5 months ago

Likely related to https://github.com/openai/codex/issues/11189 assume you're all actually seeing redirected 5.2 responses

vega113 · 5 months ago

yep, I also think I have thus issue. codex-5.3 xhigh can work 4-5 hours and not complete single task, in my experience it should not take more than 15 minutes.

BlackSwanAus · 5 months ago

Wtf is going on ?

morceautrou26 · 5 months ago

same issue. extremely slow since few days, can barely work

sunwu51 · 4 months ago

Same for me, a simple quetion for codex: how to switch reasoning level, he search about ~50 website and run ~20 bash cmds, totally cost 3min. And same question for claudecode + sonnet thinking, it return in 10s.