OpenAI has enhanced ChatGPT’s memory capabilities, particularly for users on free accounts. The improvements include a new memory architecture that builds on the previously introduced “dreaming” feature, which synthesizes information from various conversations. This update follows the initial memory feature, called saved memories, launched in April 2024, which required explicit user prompts to function effectively.
Users reported that the memories generated by ChatGPT became less relevant over time. OpenAI acknowledged this, stating that the dreaming feature was introduced to overcome these limitations. “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 said.
The newly released memory architecture is described as “significantly” more capable and compute-efficient. It will allow ChatGPT to generate a “memory summary” that users can access anytime for review and updates. Users can add or edit their information and instruct ChatGPT on when to utilize their preferences.
The roll-out includes a memory sources feature introduced alongside GPT-5.5 Instant, enabling users to see the contextual information ChatGPT used for personalization. OpenAI indicated that the enhancements will allow for greater visibility and control over memory management.
In practical use, the updated architecture will enhance ChatGPT’s ability to retain context over time. For instance, if a user discussed their camera in prior conversations, ChatGPT will tailor product recommendations accordingly. Similarly, it can inform travel plans by recalling past travel conversations. The system will also automatically update memories to prevent referencing outdated information.
This memory architecture is beginning its roll-out today for Plus and Pro users in the U.S. Notably, the efficiency improvements behind the scenes will soon enable free accounts to start recording memories using the dreaming process as well. OpenAI plans to extend these new features to users in additional countries over the coming weeks.
