OpenAI made some important improvements to the memory feature of ChatGPT, especially if you were using the chatbot with a free account. Before getting to that development, a quick recap will help set the stage for what to expect: OpenAI shipped its first memory feature in April 2024. By the company’s admission, this early implementation, then known as cached memories, was fundamental. It depended on strong signals from the user, such as directly telling ChatGPT to remember the truth. People also find that chatbot memories fade over time. So the following year, OpenAI began working on the first version of the feature that would eventually become known as Dreaming.
Dreaming runs in the background, allowing ChatGPT to combine information from many different conversations without relying on explicit commands to remember something. “Over the past year, dreaming has populated caches to improve ChatGPT’s ability to personalize feedback and de-clutter caches,” OpenAI explained. “However, historically it has never been sufficient as a standalone memory system.”
That brings us to today’s release, which sees OpenAI release what it describes as a new memory architecture that builds on the dream process to provide something “remarkable” capable and efficient computing. Now, as ChatGPT collects information about you, it will write a “memory summary” that you can read at any time. From there, you can add and update information about yourself, and tell ChatGPT if it should refer to what it knows about you and your preferences. “If you want to go into a certain area to learn more, just talk to the model,” OpenAI said. The new snapshot is designed to complement the OpenAI memory resources feature released alongside GPT-5.5 Instant. Sources allow you to see the ChatGPT information used to personalize the response, and edit or delete that context as you wish. You can see both features pictured above.
Besides providing greater visibility to the user, OpenAI claims that the new dream design is better at driving context. For example, if you’ve talked to ChatGPT about photography in the past and mentioned the camera you’re currently using, the chatbot will be able to generate personalized results the next time you ask for product recommendations related to your photography setup.
Similarly, OpenAI says the new architecture is better at tracking preferences. Say you’re planning a trip, ChatGPT will use things it’s learned from previous travel conversations to inform its responses. Using photography as an example, it can suggest a Singapore tour that includes suggestions on places where you can do street photography. To round it all off, ChatGPT will automatically update its memories as time goes by, so it doesn’t do anything like referencing a trip you’ve taken in the past as if it were coming up.
OpenAI is starting to roll out the new memory architecture to Plus and Pro users in the US starting today. Due to the improvement of the efficiency of the scenes, ChatGPT, for the first time, will start recording memories with the dream process of free accounts. As a Plus or Pro user, those enhancements will translate to ChatGPT offering a larger memory capacity. The new build will be rolled out to users in other countries in the coming weeks.
