I’m a ChatGPT Pro user and I’ve been experiencing a serious issue for almost two days.
The problem is that ChatGPT on the desktop web version suddenly behaves as if it is running in an “instant” or low-reasoning mode, even when I select the Pro / reasoning model. The replies come back almost immediately, with very little apparent reasoning, much shorter analysis, and noticeably worse quality than usual.
This is not just a subjective “answer quality” complaint. The behavior is very different from normal Pro model behavior:
- Complex coding, analysis, or reasoning prompts are answered in only a few seconds.
- The responses are much shorter and less structured than before.
- The model feels like it is silently falling back to a cheaper/faster route.
- The same account works much better on the mobile app.
- The issue is mainly happening on desktop web in Chrome.
- Switching VPN nodes seems to affect the behavior somewhat. The issue became worse after switching to a Japan VPN node, and improved slightly after switching back to a US node, but it has not fully recovered.
Environment:
- Plan: ChatGPT Pro
- Platform affected: ChatGPT web
- Browser: Chrome on macOS
- Mobile app: appears normal with the same account
- Duration: almost two days
Expected behavior:
When I select a Pro / reasoning model, the web version should actually route the request to the proper Pro reasoning model and provide normal reasoning-quality responses.
Actual behavior:
The web version often replies almost instantly and the quality looks like a fallback / instant route, even though the UI still shows the Pro model. This makes the Pro subscription almost unusable for complex tasks on desktop.
I have already contacted Support, but the replies so far have been generic canned troubleshooting responses. They did not address the possibility of a model-routing, account flag, IP risk, or web-client fallback issue.
Could someone from OpenAI please check whether there is a known routing/fallback issue affecting ChatGPT Pro users on the web, especially when VPN/IP reputation or region changes are involved?
It would also be helpful if the product could clearly show whether a request is actually being served by the selected Pro reasoning model or silently routed to a lower-latency fallback.
