OpenAI has upgraded ChatGPT’s memory system with a new architecture that the company says delivers “significantly greater performance and better computational efficiency.” The update lets the assistant summarize stored user information on demand; users can read these summaries and update the stored details at any time.
Background
An initial, rudimentary memory feature arrived in ChatGPT in April 2024. That first version removed the need to repeat certain user details in every session — for example, the model could remember that a user is vegan and avoid suggesting steak at a Sunday dinner. Users could configure what the assistant should or should not remember in the application settings.
Because that early approach had limitations, OpenAI began work in 2025 on a background process called “dreaming.” This function runs behind the scenes and allows ChatGPT to consolidate information from multiple conversations without relying on explicit user instructions to store memories.
What the new architecture changes
The newly introduced memory architecture improves how ChatGPT manages context: it can more effectively summarize and recall relevant user data and follow rules about when to reference those data. According to Engadget, users can view and update the stored information and specify when the assistant should consult those records.
A practical example: if a user previously discussed photography and mentioned which camera they use, the assistant will take that into account the next time the user asks for accessory recommendations. As a result, it will be more likely to suggest only accessories that are compatible with the user’s kit.
Rollout and availability
OpenAI has begun rolling out the new memory feature and says it will make it available in all countries over the coming weeks. The goal of the change is to provide better context retention and more efficient computation, especially when consolidating information across multiple conversations.
Why this matters
Improved memory and context handling directly affect how personalized and relevant conversational assistants can be. Background summarization combined with user controls reduces the need to repeat known details in each session while increasing transparency, since users can see and edit what the system remembers about them.
