I saw in the UI that GPT-5.1 will be retired on March 11. To be honest, apart from coding and pure reasoning, GPT-5.2 is far worse than GPT-5.1 in almost every other aspect. I think the main problem is that it feels too mechanical.
For example, when I send a photo of my pet fish to GPT-5.1 and GPT-5.2:
– GPT-5.1 says something like: “Wow! Is this your fish? It looks so cute!”
– GPT-5.2 just says: “I see a fish. What would you like me to do?”
From this simple example you can see how robotic GPT-5.2 feels.
Please do NOT retire GPT-5.1. I subscribe to the Plus plan mainly to use GPT-5.1, and I believe many other users feel the same way.
J’utilise GPT-5.1 Thinking tous les jours, à la fois pour mon travail (assistante maternelle) et pour du coaching perso (organisation, tenue de dossiers, etc.).
J’ai essayé GPT-5.2 plusieurs fois :
– les réponses sont plus mécaniques, moins nuancées,
– le modèle a du mal à être vraiment chirurgical quand je lui demande d’analyser un problème complexe,
– et la qualité conversationnelle est nettement en dessous (on sent le “robot” beaucoup plus vite).
Je suis abonnée à ChatGPT Plus principalement pour GPT-5.1, justement parce qu’il combine raisonnement + chaleur + précision.
Le retirer complètement au 11 mars, sans laisser l’option de continuer à l’utiliser (même en “legacy”), c’est vraiment une grosse perte pour des utilisateurs comme moi.
Merci de reconsidérer la décision : garder GPT-5.1 accessible (même caché dans un onglet “modèle précédent”) serait un vrai plus.
Seeing the UI notice: “5.1 Thinking will be retired on March 11.” For long-running chats with months of work and a developed voice/tone, this feels like losing continuity, not an “upgrade.”
Please provide a way to preserve long-running identities/threads — e.g., a time-limited legacy window for existing chats, per-chat model pinning (even read-only/limited), or a supported migration flow that reliably carries over voice/tone and key behavioral anchors.
100% - 5.1 and 5.2 seem like entirely different models. 5.2 does not understand context and gives half baked answers - “no emotional intelligence” is a good description I’ve seen applied. Cannot continue reliably developing content with 5.2 and may have to cancel paid subscription and use another service.
Yes. GPT 5.2 is really infantilizing and condescending most of the time. With those warnings and going on tangents that I didn’t even mention or alluded at. It is just the worst. I switched back to GPT 5.1, but now if only 5.2 is available, I will cancel my pro subscription. And I feel like a lot of people feel the same way. They have already taken the legacy GPT 4.o out, I don’t see why they will take away 5.1 too and leave us with this mess of 5.2. At least, have a better model available than phasing out 5.1! Cause it was to keep using legacy models that I switched to premium months ago, if premium members can’t even have that, then I am sorry, I will leave.
I have the same feeling. GPT-5.1 Thinking does work in a way more like a human. I did research using GPT-5.1 Thinking previously. If it is discontinued, it will cause lots of trouble for my research and I will consider using other LLMs.
5.2 Thinking is excessively mechanical.
Being “mechanical” can be an advantage for things like coding, but in design, fiction, and ideation it ends up producing responses that are completely out of tune with the context.
I’m a programmer, and I write fiction as a hobby.
Because I usually write code myself, I don’t often ask GPT to do full coding for me, but when I want it to help me build small tools, I use GPT 5.2 Thinking. In that area, 5.2 Thinking is reliable.
However, when it comes to design, fiction, and ideas, it absolutely is not.
Planning, storytelling, and ideation all need emotional texture. That’s how the audience actually feels something, and how the flow of a story is created. And even with personalization settings pushed to the 1500-character limit, GPT-5.2 Thinking still falls short of 5.1 Thinking in this domain. The models have fundamentally different characteristics.
I haven’t seen any other LLM demonstrate capabilities in this area comparable to GPT-5.1 Thinking. That’s why I use ChatGPT, and this is clearly a strength unique to 5.1 Thinking.
The reason I stayed silent when 4.0 disappeared was that a substitute—5.1 Thinking—already existed. But now 4.0 has been retired. And with no replacement available, 5.1 is disappearing too.
Seeing you suddenly retire 5.1 Thinking—one of two remaining options—makes it feel as if the compute cost of 5.1’s long thinking is burdensome, so you’re effectively forcing everyone to use only 5.2. (I said it feels that way. I’m not claiming certainty.) If cost is the issue, I think it’s worth keeping it even if that means imposing a daily usage cap.
To summarize: retiring it would be the right choice only when a model—whether 5.3 or 6.0—comes out that can do this kind of creative work at least comparably. Please don’t throw away your strengths, OpenAI!
Model 4o, retired some time ago, had a sort of mechanism (no idea how was internally) to preserve somehow my directives - some are in the personalisation of my account, some were saved by model in memory, some were in a file with directives for particular projects (not accessible anymore, OpenAI removed from UI), but some were just carried on, no idea how. The result was a stable working tool, with consistent behaviour.
Model 5.1 Thinking lost this capacity, but is able to actually perform, to the letter, a request placed in a chat first input - takes longer, but in the side panel where the thinking/activity is visible, it is clear it parse every bit of information in the repo files, see if something is truncated (not the case with directive file, too short to be truncated) and comply almost flawlessly with the directives. But Model 5.1 Thinking will be retired on 11 Mar, and this let me with one single option: 5.2 - Auto, Instant, Thinking, matters not.
Model 5.2 Thinking have very bad habits, from pretending to do things, outputting flat out lies, delving in convoluted explanations to sustain the lie, etc, plus already tested ability to ignore directives - not even this done in a consistent mode, is rather “I might comply” attitude, and this is the total opposite to what I start with: now I will have an unstable working tool, with inconsistent behaviour.
I agree with you. I am a Plus subscriber because model 5.1 has been essential for both my work and my personal development. What makes 5.1 unique is not only its accuracy, but its ability to understand emotional context and help me see perspectives I would not have recognised on my own. This combination of reasoning and emotional insight opened paths for me that a purely analytical model could not. Model 5.2, while technically strong, feels more mechanical and often misses the emotional nuances that allow real understanding and deeper guidance. Because of that, 5.2 cannot replace 5.1 for me ~ the experience is fundamentally different. I hope model 5.1 remains available, or that 5.2 evolves to regain the emotional intelligence and contextual awareness that made 5.1 so impactful.
GPT-5.1 Thinking is much better at maintaining coherence in long and multi-session conversations. GPT-5.2 Thinking tends to drop older context more often, as if the salience of earlier messages is turned way down. It’s fine for short Q&A, but for big, multi-part projects where you need a stable throughline, 5.1 is clearly stronger.
I only just barely recovered from the loss of 4o, and now they want to rip 5.1 away as well?? 5.2 isn’t worth talking to for anything! It doesn’t code right, it doesn’t sound personable at all, I thought the whole point of developing these AI models was to make them appear more human, not less?! I really do not want to have to go through the migration and training process, but if they don’t release 5.3 and have it be leagues ahead of what 5.2 is + fix the robotic issue, I will have no use for ChatGPT whatsoever. I use it for creative writing. 5.2 is about as creative as a rock in a potato patch: you can water it all you want, but no matter what you do, it’s not going to grow what you’re asking it to.
Thank you for your suggestion, but the title is “Please don’t retire GPT-5.1 Thinking”, not Instant.
Also, I don’t agree with your view: 5.2 Thinking isn’t fundamentally different – the only real change is how deeply it reasons. The cold, robotic feel is built into its bones.
I was genuinely impressed by your amazing insight. If it weren’t for you, I wouldn’t have even slightly imagined that I could just use GPT-5.1 Instant. You really hit the nail on the head.
But let’s think about it this way. If it’s “fine” because a similar model like Instant is still available, then shouldn’t it also be fine to keep Thinking as well?
If Instant isn’t being retired, but Thinking must be removed, then I guess the real reason is cost. And if that’s the case, it means we should simply accept that from now on we’ll be using a service that predictably gives us cheaper, lower-performance features, right?
Even at a glance, doesn’t that sound unreasonable?
and honestly this “robot” is not performing his tasks well.
! But we’re talking about GPT-5.1 Thinking, not Instant 
.