My issue was finally resolved after a month, and it really does appear to be IP-related. This seems to happen mainly to users who connect through VPNs or similar tools. In these cases, ChatGPT appears to silently route the Pro model to a Mini model instead. Because of this, contacting customer support is unlikely to solve the problem.

If you want to get your account back to normal, here are a few things you should know:

Stop repeatedly reporting the issue to support.
Reporting it to customer service is basically ineffective. They simply do not seem to have the ability to resolve this particular problem.
Understand how the issue works.
Even when two accounts use the exact same network node, one account may have this problem while another continues to work normally. However, more and more users seem to be experiencing it.
Switch to a different network node.
Try to find a clean, high-quality IP, preferably a genuine residential broadband connection. Make sure it is actually residential and not just advertised as such.

If you cannot find a reliable residential IP, I would suggest avoiding heavily used locations such as the United States, Hong Kong, and Japan — especially the United States. Instead, try locations that appear to be less heavily used, such as Canada, France, or South Korea.

Switch and wait.
After switching to a new network node, your account may not recover immediately if you had previously been using a problematic IP. You may need to wait one or two days.

If the account still has the same problem after two days, the new node is probably also problematic, and you may need to keep looking for a better one.

I have two Pro accounts, and both returned to normal after I switched to a new network node. I had also left them unused for several days beforehand.

I then ran an experiment with one of the accounts that had already recovered. I switched the network back to the old problematic U.S. node, and the issue returned almost instantly: the Pro model was routed to Mini again.

Even after I immediately switched back to the new node, the problem remained. The account did not return to normal until about a day later.

Based on this repeated test, I am 100% convinced that, in my case, the issue was caused by the IP/network node.