Issue: My ChatGPT Pro subscription has been serving GPT-5.3 Mini for every single response for over 48 hours, despite Pro Extended mode being enabled. “Use Pro mode” on individual responses also has no effect.
What I’ve ruled out:
Not a browser/cookie issue — persists in Chrome Incognito with a
Not a DNS or IP leak — verified a clean Singapore exit node via ipleak.net
Not a visible rate limit — no reset timer appears anywhere in the interface (model picker, settings, or inline)
Support status: Case #10179682 escalated to a human specialist, awaiting response.
Question for the community: Has anyone else experienced Pro model routing being locked to Mini for multiple days without a reset timer? Did it resolve on its own or require support intervention?
I reported the issue via email. I shared screen recordings explaining the problem.
They are looking into it.
It has been almost 5 days now; I am still facing this issue.
This is really frustrating. You do not expect this service for the price you are paying.
I have the same problem. It has persisted for 8 days. I had a similar experience before, but simply replacing the output solved the issue. This time, no matter what output I change, it won’t work. Opening the pro mode immediately gives an answer without any time for thinking. After checking, it was found that it automatically dropped to the 5.3 mini model. But the thinking mode is functioning normally. But did I really pay $200 per month for the thinking mode?
Just reported, so far no response has been received. As you said, it won’t be able to be completed quickly. I also saw that others said they didn’t receive a response after 5 days of feedback. Spending the most money results in the worst treatment. , PERFECT!
I’m back online. After various attempts on Saturday night and not using the web interface on Sunday, I have been using Codex today. When I checked again at work today, it was back to normal. I have no idea what happened. During this period, I received a reply from the OpenAI official, but the gist of the reply was that we can’t contact you via the web, but only through email. Reading this email, you can go to the “OpenAI Help” webpage at the bottom right corner to start a conversation. However, when I tried to initiate the conversation, it was still the AI chatting with me.
Exactly the same! I bought a $200 subscription. After logging in at home and at work, it seems to have detected that I was abusing the service and downgraded all services to GPT 5.3 mini. I logged out of all accounts and waited 48 hours, and it seemed to return to normal. But today I logged in at work again and the problem came back. We are paying so much money but getting service like garbage. Worse, we’re not getting any effective help or response—this is insane!
The same problem too. It has persisted for at least 8 days from 6.15-6.22 now! The same that ‘The thinking mode is functioning normally, but Pro routed to 5.3-mini.’ On June 16, I fully confirmed that the mini model was not caused by any of the issues mentioned in the Help article. I thought it was because I had used up my overall quota, so on June 18 I paid another $200 to renew/top up my plan. Now I feel like a complete clown.
The latest situation is as follows: Since I have three devices, the company computer, my computer, and my iPhone, the first to take effect was the company computer, and it restored Pro Thinking, but it only lasted for two days. Today it failed again, and the home computer has never been restored and remains in the Pro mode without thinking. However, the phone has always been working fine. After the message is sent from the phone app, it directly enters the Pro thinking mode in one second. I don’t know why.
As for your question, my Codex app is functioning normally. I stopped using the web end, and I didn’t use the CLI. It seems that the risk control would automatically be lifted within 24-48 hours, but now I seem to be more accustomed to using the Codex app to issue commands directly in the target mode instead of generating commands through the web end and then giving them to Codex. But this is really very very annoying. I bought the service but can’t use it. Why are they doing the risk control like that, just like that damned Anthropic!
