OpenAI has introduced a new memory system for ChatGPT, designed to enhance its ability to retain and utilize user context effectively. This update, called Dreaming, aims to address issues of memory staleness and scalability for the millions of users of ChatGPT. The feature is now available to Plus and Pro users in the US, with broader rollout planned for other regions and user tiers in the coming weeks.
The new memory architecture builds on previous updates that introduced a background process for curating memories. Initially, saved memories were limited to specific cues and could become outdated or irrelevant over time. In 2025, OpenAI introduced Dreaming, which allowed the model to automatically synthesize memory from chat history, improving the freshness and relevance of context. However, it was not sufficient as a standalone system. The latest iteration, Dreaming V3, is a more capable and compute-efficient memory architecture that integrates with saved memories to provide a step-function improvement in personalization and context retention.
OpenAI emphasized that the new system enables ChatGPT to maintain a more accurate and up-to-date understanding of user preferences and constraints, which is essential for delivering meaningful and consistent assistance over time. The memory summary page now provides users with an overview of what ChatGPT knows about them, allowing for easy updates and customization of memory settings.
Source: openai
