OpenAI is introducing a new version of ChatGPT memory designed to keep track of user context more effectively over long periods. The company said the update improves how ChatGPT remembers preferences, projects, and constraints by automatically synthesizing information from past conversations into a fresher summary.
The rollout begins today for Plus and Pro users in the United States, with access planned for additional countries as well as Free and Go users in the coming weeks. OpenAI described the feature as a more scalable approach to memory, aimed at handling the size of ChatGPT’s user base and the challenge of keeping memories accurate over time.
OpenAI says the revised system builds on a background process it previously referred to as “dreaming.” Rather than relying only on explicit prompts such as asking ChatGPT to remember a fact, the new system can review chat history and distill useful context automatically. The company says this helps ChatGPT surface information that appears naturally in conversation and reduces the need for users to repeat themselves.
The memory update is intended to improve three core areas. First, it should carry forward useful details from one chat to another. Second, it should respect user preferences and constraints, such as dietary choices or other recurring needs. Third, it should stay current as circumstances change, so old information does not linger after it becomes outdated.
OpenAI says earlier memory tools had limits. The original saved memories feature, launched in 2024, depended on direct instructions and specific cues to decide what to keep. That made it useful for isolated facts, but less reliable for longer-term context. The company later introduced an early dreaming system in 2025 to let ChatGPT reference broader chat history in the background, but said that version was not enough on its own.
The new architecture is meant to be more capable and more efficient, while addressing problems such as stale or incorrect memories. OpenAI says the system is now able to synthesize memory at a larger scale and with better freshness than the earlier approach.
OpenAI said the memories generated by the system will be visible through a memory summary page. That page is meant to give users a quick view of what ChatGPT believes it knows about them. From there, users can correct information, add new details, or tell ChatGPT which topics should come up and when.
If people want to explore a specific topic in more depth, OpenAI says they can continue the conversation directly with the model.
The company illustrated the upgrade with examples showing ChatGPT responding more specifically when it has long-term context. In one example, the assistant used details from prior chats to give product advice tailored to a user’s camera setup. In another, it used remembered travel preferences to help with a trip plan that matched the user’s interests.
OpenAI says those kinds of responses are meant to show the practical benefit of memory. Rather than starting over each time, ChatGPT can use a stored summary of relevant context to be more helpful across multiple conversations.
The memory update is part of OpenAI’s broader effort to make ChatGPT more personalized over time. By moving from simple saved facts to a more dynamic background summary, the company is betting that the chatbot can become more useful without requiring users to restate the same information again and again.