OpenAI is on a mission to make sure you don’t have to re-educate the AI every time you log on. This latest upgrade lets ChatGPT be a steady companion rather than something that resets with each session. You get tailored answers and quicker results, with less repetition, if you’re a regular user.
Why this is a big deal
Think of it as moving from a one-off tool to an assistant that adds up over time. Because the chatbot can hold on to context, you don’t have to put in the work to restate your preferences or set the scene when you need some help. It just makes for a stickier experience.
You used to have to tell the earlier versions of ChatGPT to put things in its head, and some of those would go stale. OpenAI’s new method is designed to learn from a wider range of conversations and be more discerning about what to keep, so you don’t run into those hiccups and the output stays on point.
More control, less friction
To go with the learning happening under the hood, there is now a bit more visibility. A new summary page will show you the kind of things ChatGPT has picked up on. You can have a look, put in a correction, or let the AI know when to bring a subject up.
And if you want to fine-tune a particular piece of info, you can just have a word with the model. It’s about having a mix of hands-off learning and being in the driver’s seat when you are.
What you’ll see
Here is the gist of what OpenAI is putting out with this version:
– No need to prompt it to carry over context from one chat to the next.
– Some ‘dreaming’ in the background to sort through details.
– An answer to the usual problems of staleness and scale.
– A one-stop page to review your memory file.
– In the U.S. for Plus and Pro users as of today.
– Free and Go users will be in on it in the weeks to come.
How it works
OpenAI is calling it a more efficient architecture, and it’s all built around ‘dreaming’. While you’re not looking, the process is mulling over your various conversations so it has the right context ready for you when you come back.
They say it puts to rest three old headaches: whether the memory is fresh, right, or can be scaled. By making better links between what you’ve said in different sessions, the model can hold on to the long view without getting in the way.
Take a trip to Singapore for instance. If you’ve made it known you like wildlife photography, a good hotel A/C and a quiet dinner, the next time you plan an itinerary it should be in there. No starting from scratch.
A brief history
We first saw this in April 2024, when you could have ChatGPT put something in its memory for later. It was fine, but you had to be explicit about it, which made for a stiffer conversation.
Then in April 2025, it could pull from old chats even if they weren’t on the saved list. OpenAI dubbed the engine behind it ‘dreaming’ – a way to synthesise a better recall from many talks at once.
What we have today is an evolution of that. The system is more adept at sifting and using context, and according to OpenAI, you’ll find the whole thing is more dependable and doesn’t repeat itself as much.
When and where
If you’re a Plus or Pro in the U.S., you can use it now. The plan is to open it up to other parts of the world and to the Free and Go tiers in the near future.
For you, it means you won’t be left cold when you start a new chat. For OpenAI, it’s about the long haul. They want a product that gets to know you, and in doing so, becomes an assistant in the truest sense.
