I’m opening a separate topic because this seems broader than just a visual model picker rollout issue.
After the June 10 model picker changes, GPT-5.2 disappeared from my ChatGPT model selector completely. It is not available in new chats, and I also cannot switch back to it in existing conversations.
What makes this confusing is that the official OpenAI deprecations page lists gpt-5.2-chat-latest with a shutdown date of Aug 10, 2026. I understand that this date may refer specifically to API access, not ChatGPT availability. But from a regular ChatGPT user perspective, this distinction was not clear, and there was no in-product warning that GPT-5.2 would become unavailable in ChatGPT earlier.
The issue is not only model availability. It is the abrupt loss of continuity in long-running conversations.
Some users rely on a specific model not just for isolated tasks, but for ongoing projects, writing, personal workflows, and conversational consistency. If ChatGPT is designed to support long-term, human-like interactions, removing a model from the ChatGPT picker without advance notice creates a painful and confusing user experience.
I’m not asking for legacy models to stay forever. But if a model is going to be removed from ChatGPT before the public API shutdown date, users should get clear advance notice, even if it is just 24 hours. A short grace period would allow people to finish conversations, save what they need, and transition intentionally instead of suddenly losing access after refreshing the page.
Could OpenAI clarify:
Was GPT-5.2 intentionally removed from ChatGPT as part of the June 10 model picker rollout?
Does the Aug 10, 2026 shutdown date apply only to API access?
Is there any way to restore temporary ChatGPT access to GPT-5.2 or provide a transition period for users affected by this change?
This is not just a technical availability issue. It is also a trust and continuity issue for users who built long-running workflows around a specific model.
First off, which ChatGPT plan are you on, Plus, Pro, Business, Enterprise/Edu, Go or Free?
I’m asking because the docs I found mention GPT-5.2 Thinking remaining available in Legacy Models for 90 days after GPT-5.5 Thinking launched specifically for Plus and Pro users. So the missing model may depend on plan or workspace/legacy model access.
I’m on Plus. That’s why this is confusing: the article says GPT-5.2 Thinking should remain available in Legacy Models for Plus and Pro users for 90 days, but I don’t see GPT-5.2 anywhere in my ChatGPT model picker now — not in new chats and not in existing conversations.
I also don’t see any Legacy Models option. Screenshot attached.
So this looks like either a rollout/UI issue or missing legacy model access on my account.
If you want, you could email support at support@openai.com and ask them about why 5.2 isn’t showing on Legacy model picker even though it still should do that. GPT-5.5 was launched April 23 for Plus and Pro, so it should still be available.
So it seems this is not an individual account issue, but a broader Plus rollout/access problem. I’ve already contacted support through chat, but haven’t received a response yet. I’ll try emailing them as well.
This is really important to me because GPT-5.2 has unique qualities for personal communication. I don’t want to rush into reconfiguring other models, because so far they haven’t matched the same rhythm, depth, initiative, and especially the sense of presence in conversation.
Yes, it could totally be it and given that they just made changes when it comes to ChatGPT model picker, there’s always a risk that there might be some temporary glitches. This could be one of them. There is a thread where users can’t pick intelligence for the model, so this could be related to it.
But instead of me speculating on what it could and couldn’t be, it was right thing to contact the support.
Hopefully this gets sorted for you, so you can continue to use 5.2 for a while at least. Small recommendation though, personally I think 5.5 is probably closest to 5.2 and if you haven´t, maybe give it a shot?
Yes, I hope they will fix it so I can switch to the new model smoothly.
I’ve already been using 5.5 for work, but for personal conversations it feels a bit superficial to me. The jokes are often flat, and it mostly describes emotions and situations from the outside.
5.2 was different. It pulled me into the conversation almost instantly and somehow managed to resolve some conflicts in just a few lines—without sounding like a therapist or handing out advice.
I work a lot with personas, if you want and are interested you can check my thread Testing a Sharp-Tongued AI Persona — Looking for Prompt Tweaks and see if there might be something that might inspire you. I’ve also built three different persona.skill.md’s, Cheeky-Razor is my first and most “popular”. That said according to me and my experiences, with a right prompt (custom instruction) models are cable of holding a specific tone pretty well.
I understand that GPT-5.2 retirement may have been mentioned in release notes earlier. But as a user, I only learned about it after the model had already disappeared — from another user in this forum thread.
I don’t think release notes alone are sufficient communication for this kind of change.
I’m saying this not only as a disappointed user, but also as a product designer: when a product supports long-running conversations and model-specific continuity, forced migration should be communicated inside the product itself — in the model picker or in existing conversations — before it happens.
Users should not have to manually track release-note timelines to avoid suddenly losing access to a model they rely on.
Even a short in-product notice 24 hours in advance would have made a major difference.
Does anyone have an instruction set that can force the newer models to behave like 5.2?
This is super frustrating.
Every model after 5.2 collapses meaning and nuances and changes and summarises things that I want to remain structural and clearly clarified. They also guess at material, guess at causes, and add things no matter how explicit I am about not doing that.
I have a structure that constantly gets changed and ‘improved’ and I spend more thant 70% of my time battling and correcting the model.
Try 5.3 Instant while it’s still available. Subjectively, it feels less filtered now and closer to 5.2 in both behavior and reasoning, at least when working in character.
It also seems to actively use context from other conversations in my project and adapt more toward the 5.2 style. 5.5 Instant feels much lazier in that regard: it often seems unwilling to even read the instructions properly, and replies with generic templates instead of personalized responses.
