OpenAI is enhancing ChatGPT’s memory feature, particularly for free account users, introducing a new architecture that significantly improves its personalization capabilities. The initial memory system, implemented in April 2024, required users to prompt the chatbot to remember information, leading to concerns about memorized facts becoming irrelevant over time.
To address the shortcomings of this system, OpenAI developed a feature called “dreaming,” which synthesizes information from various conversations without explicit instructions from users. “Over the last year, dreaming supplemented saved memories to create a step-function improvement in ChatGPT’s ability to personalize responses and offset the staleness of saved memories,” OpenAI stated. Despite these advancements, dreaming was not sufficient as a standalone memory system.
The latest release introduces a new memory architecture that builds on the dreaming process. Now, ChatGPT will generate a “memory summary” that users can access and edit anytime. This summary allows users to add personal information and instruct ChatGPT on when to reference their preferences. OpenAI noted, “If you want to drill down into a particular area to learn more, just chat with the model.”
This release also includes a memory sources feature introduced alongside GPT-5.5 Instant, enabling users to see, edit, or delete the information ChatGPT used for personalization. OpenAI emphasized that the new dreaming architecture is better at carrying forward context from past conversations. For instance, if a user discussed photography and mentioned their camera, ChatGPT will provide tailored product recommendations in future interactions.
ChatGPT will also track user preferences more effectively. For example, if users are planning a trip, it will incorporate insights from past travel conversations to enhance its responses, such as suggesting a Singapore itinerary focused on street photography. Additionally, the architecture will automatically update its memories over time to ensure relevance.
OpenAI is rolling out the new memory architecture to Plus and Pro users in the U.S. starting now. Free account users will soon benefit from the memory recording capabilities through the dreaming process due to efficiency improvements. Enhanced memory capacity will particularly benefit Plus and Pro users. The rollout to users in other countries is expected in the coming weeks.
