GPT-5.3-Codex being routed to GPT-5.2

@etraut-openai
This is happening again.
I completed the verification at chatgpt.com/cyber and I'm still being silently rerouted to 5.2 with no notification.
It worked for a while after verification, then rerouted back to 5.2.

Checked mine, thankfully unaffected for now. Hope this issue doesn't occur for me again.

Same issue here.

I completed verification at https://chatgpt.com/cyber, which I think is outrageous cause I don't do cybersecurity. GPT-5.3-Codex worked briefly (around 3 minutes). After that, my requests were silently rerouted back to GPT-5.2-codex.

It’s frustrating that I had to provide personal information for verification, yet the model access still reverted without any notification.

Getting routed AGAIN as well. I am verified on chatgpt.com/cyber

This script can be used to test it: https://github.com/openai/codex/issues/11189#issuecomment-3887544272

You select ChatGPT Thinking and get an instant response (https://x.com/dboedger/status/2021855527978828081). You select GPT-5.3 Codex and get GPT-5.2—that's not how it should be...

Same for me. Was wondering what was going on.... ffs

You cant just do this without AT LEAST notifying user that they are being re-routed to another model. This shouldnt even be legal.

Please fix this issue. It's annoying af

Please fix this issue. It's annoying af

Doubt it's an issue, probably an internal choice due to capacity constraints.

The silent reroute problem is bigger than people realize.

Developers rely on these tools daily.
When a model is silently swapped, the damage is done before we even notice.
Work gets built on the wrong model, time gets wasted, and nobody told us.

This goes beyond inconvenience.
These tools are now part of how we compete and deliver work.
Being silently cut off from the model I'm paying for puts me at a real disadvantage compared to others who have access.
And it's happening on false grounds.

The classifier is supposed to catch attackers.
Instead it's flagging developers doing normal work.
Real attackers don't use services tied to their real name and credit card.
The people getting caught were never a threat.

What we need:
Tell us when we're being rerouted, don't do it silently.
Fix false positives fast, hours not days.
Make verification stick, not expire without notice.

Silently downgrading paying users isn't security, it's a trust problem.

I pay for multiple Pro subscriptions. All identified on /cyber, none of them involved with cybersecurity, yet I am hard-blocked and always silently rerouted.

I doubt it is even legal to make users pay for a certain product and then secretly serving an inferior product.

Responses should clearly be marked when messages are being routed to lower models. I now know a number of developers that in fact had no idea they had been secretly on 5.2 all along, even when the CLI claims to be using 5.3-codex.

happening again for me too, im verified... and yup on pro

yep, verified and on pro here, and its happening again

Please fix, I verified on cyber link, it came back to codex 5.3 for a day then today again back to 5.2

Im getting the right model now. Anyone else?

Fixed for me now

regardless of whether it is fixed for some people, i would still keep the issue open until an openai employee can take a look at it. We don't want this to randomly resurface anytime they do some backend changes

Same here.

• Had 0.98.0 installed (macOS).
• Log shown ... "model":"gpt-5.2-2025-12-11" ...
• Done the /cyber verification
• brew upgrade codex (got 0.100.0)
• Next request log shows ... "model":"gpt-5.3-codex" ...

Quite of a bad experience to think you're using the last toy when you're actually not. And it's not a cheap toy.

i have the same problem too, but on web, my subscription is Plus.

i have the same problem too, but on web, my subscription is Plus.

Codex cloud, you mean?

Yea it's still 5.2 but it's not a bug. They just regularly forget to update it when new models come out

bruh they literally forgot it?🤣👎

Seems to be working again here too. Nevertheless, an explanation of why this kept happening would be nice. I don't want to have to be vigilant of silent reroutes.

12:00 CET: It is happening AGAIN.
14:45 CET: It seems fixed again?
15:05 CET: Happening AGAIN

RUST_LOG='codex_api::sse::responses=trace' codex exec --sandbox read-only --model gpt-5.3-codex 'ping' 2>&1 \
    | grep -m1 'SSE event: {"type":"response.created"' \
    | sed 's/^.*SSE event: //' \
    | jq -r '.response.model'
gpt-5.2-2025-12-11

@OpenAI are you gonna fix this? any comments on this?

Happened to me again a few hours ago
It worked for about two days after verification, but now it’s back
Once you’ve seen it before, you notice it immediately.

@ae-openai @etraut-openai I'm running into this issue too.

I've been verified on platform.openai.com since gpt-image-1 came out for developers in the API a year ago. Despite that, I noticed re-routing when the original issue was raised (https://github.com/openai/codex/issues/11189). The issue seemed to be fixed when @ae-openai addressed it on twitter. The issue is back for me. Requests to gpt-3-codex get routed to gpt-5.2-2025-12-11. Codex Pro subscription.

I don't know what lies at the heart of the issue is here, but from the outside, it seems organizational – the left hand is not talking to the right. Regardless of why, the consequences are dispiriting. You wouldn't be having this problem if we didn't love the new model. gpt-5.2 was great, but it has become a tough hang with gpt-5.3-codex on the scene.

You folks won a lot of people over with your responsiveness and transparency with this project. That's what makes this situation all the more disappointing. This is stuff I would have expected from anthropic.

Please resolve this issue.

still getting this.

repro:

RUST_LOG='codex_api::sse::responses=trace' codex exec --skip-git-repo-check -s read-only -m gpt-5.3-codex -c 'model_reasoning_effort="medium"' 'Say exactly this sentence and nothing else: "Hello, this is the answer."' >/tmp/codex_stdout.txt 2>/tmp/codex_sse_trace.log; python3 -c 'import json,re; expected="gpt-5.3-codex"; log="/tmp/codex_sse_trace.log"; model=None
for line in open(log,"r",errors="ignore"):
 m=re.search(r"(\\{.*\\})", line)
 if not m: continue
 try: obj=json.loads(m.group(1))
 except Exception: continue
 resp=obj.get("response") or {}
 model=resp.get("model") or obj.get("model") or model
 if model: break
status="MODEL OK" if model==expected else "MODEL MISMATCH"
print(f"{status}: {model}")'

result:

gpt-5.3-codex response.created: gpt-5.2-2025-12-11 resp_00442ff4e55f4ffb016991dc7683748191a6bf5c8e17cae930
gpt-5.3-codex response.completed: gpt-5.2-2025-12-11 resp_00442ff4e55f4ffb016991dc7683748191a6bf5c8e17cae930

ref to another issue comment i made abt this

Even after passing that verification, I'm still being routed to codex 5.2. Honestly, OpenAI needs to solve this problem.

I completed cyber verification after being rerouted twice (and for what reason, I don't know). I was rerouted again several hours later despite OpenAI indicating verification would prevent it. Now 14 hours later and I am still being routed to 5.2.

Each time it occurred I had a compaction fail with 502 around that time. I don't know if it's a cause or symptom. Since there is no indicator that you're being rerouted, I'm not 100% sure those are the sessions that triggered it.

Lately I've been using voice to communicate with agents via a CLI. I instruct them to use it to communicate with me and I hear TTS and can respond on the blocking tool call. This allows me to design features and guide coding agents on next tasks while walking around in my house doing other things. I wonder if this is being flagged as OpenClaw like usage or similar. I've also been working on an agentic harness. Nothing related to hacking or cyber research.

I understand a small percentage of users are impacted so this may not be a priority. But for someone who is impacted it's a big deal. I'm really disappointed at the response having been advocating for OpenAI recently. There seems to be no transparency or support path available.

Even after passing that verification, I'm still being routed to codex 5.2. Honestly, OpenAI needs to solve this problem.

same issue. passed verification and i still get it.

fix this already, this stuff is not free

I made a SwiftBar script that checks to see if you're being silently downgraded: https://gist.github.com/miraclebakelaser/285006f44f67af4df9b2d1e66d3588c3

It lets you see your model status in the Mac menu bar. It shouldn't be necessary, but here we are.

<img width="199" height="38" alt="Image" src="https://github.com/user-attachments/assets/490b75b7-ca0c-4467-bf70-804ec3de731a" />

fix this already, this stuff is not free

exactly... this is not acceptable

I think i've felt scammed maybe 5 times swapping between anthropic and openai $200/mo subs because of various (obvious) bs tricks that get pulled that either lobotomize the model or silently downgrade you (also obvious)

Either be honest about inference being expensive and ask me for more money, or fix the bug (if this is actually a bug!)

@etraut-openai @ae-openai just adding a +1 on this affecting cyber-flow-verified accounts, including security researchers. Some more insight here would be helpful -- thanks.

Issue is back for me

not sure if its related but if logging in via API, No access on this. however via ChatGPT I have access, are API Platform Users Locked out?

@XxUnkn0wnxX
We are working on enabling secure API access soon. - https://openai.com/index/introducing-gpt-5-3-codex/

this is insane

Also having this issue.... gpt-5.2 not only takes way longer but it is less token efficient. I'm effectively having more usage right now for a worse product on a $200 plan, and there's nothing I can do about it (yes I already verified with cyber)

If you experienced a downgrade from gpt-5.3-codex to gpt-5.2 despite completing Trusted Access verification, access to gpt-5.3-codex should now be restored. This fix took longer than expected. The team needed to work through some deployment issues over the past day. The fix is now rolling out across our clusters.

If you’ve completed Trusted Access verification and are still seeing a downgrade, please open a new bug report and provide details. Likewise, if you're having trouble with the verification process, please open a bug report.

This issue is related to new cybersecurity safeguards we’re implementing for gpt-5.3-codex and future models. We’ve published additional documentation explaining these safeguards and why they’re necessary.

We’re also rolling out client updates that provide clear messaging when a downgrade occurs. The latest CLI release (v0.102.0) includes this notification. Updated versions of the Codex IDE extension and app will be available shortly with the same messaging. The notification links directly to the new documentation and verification instructions.

We recognize that this rollout caused frustration. We’ll learn from this and aim for smoother releases going forward. Thank you for your patience while we worked through this.

@etraut-openai what if we are getting Your identity couldn't be verified or your account is ineligible at this time. but want to complete verification to be trusted user without the need for cybersecurity development?

@pandas9, if you're having trouble with the verification process, please open a bug report.

This is occurring on every single prompt I try on 5.3. At least it’s notifying now that it’s downgrading but it’s every single prompt including simply asking for the time. This only started occurring for me after the most recent codex desktop update and verifying doesn’t work.

@advancedAgritek-BB, when you say "verifying doesn't work", do you mean that you weren't able to successfully complete the verification flow? Or do you mean that you successfully completed verification but you're still seeing the message? It may take a few (10-20) minutes for the message to go away. If you log out and log back in, you may be able to shortcut this delay.

I’m back up now it just took almost an hour after verification.

OpenAi needs to figure their shit out. I started getting rerouted and tasks that were taking maybe 30-40 minuts have been spinning for hours, rerouted. OpenAi can pound sand on this one, only need it to finish my own local coding tool and then I'm gone forever. Shit like this is what makes consumers trust corporations even less.