OpenAI expands ChatGPT memory with new synthesis system
OpenAI said it is rolling out a more capable memory system for ChatGPT that can learn user preferences, projects, and constraints across conversations. The company said the update focuses on freshness, correctness, and scalability, and is available now to Plus and Pro users in the US.
OpenAI said it is beginning to roll out a new memory system for ChatGPT that is designed to make the assistant more useful over long periods of use. The company said the update improves how ChatGPT synthesizes memory so it can keep track of preferences, projects, and constraints while reducing problems with stale or incorrect memories.
According to OpenAI, memory is what lets ChatGPT carry context from one conversation to the next instead of starting over every time. The company said that has become a central part of the ChatGPT experience over the past two years, because it helps the model understand a user’s needs well enough to support ongoing work.
The update is available today to Plus and Pro users in the US, with rollout to more countries and to Free and Go users planned over the coming weeks, OpenAI said.
OpenAI described the new system as a more capable and compute-efficient memory architecture built on top of “dreaming,” its term for a background process that automatically curates memories by referencing chat history. The company said this approach is meant to address three challenges it sees at scale: freshness, correctness, and scalability.
The memory feature first launched in April 2024 as saved memories. At that stage, users could ask ChatGPT to remember specific information, but the company said the system depended on strong cues and only wrote memories during the conversation. OpenAI said that could leave ChatGPT acting like it had taken notes but still forgotten anything not explicitly written down.
In April 2025, OpenAI added the first version of dreaming, allowing ChatGPT to reference chat context outside the saved memories list. The company said that version helped the model learn from many conversations and synthesize memory state in the background, which improved personalization and reduced staleness. Still, OpenAI said it was not enough on its own.
The company said the memories synthesized by the new dreaming-based system can be reviewed on a memory summary page. That page lets users see what ChatGPT knows about them, correct or update details, and set instructions about what topics the assistant should bring up and when.
OpenAI said it evaluates memory on three main goals: carrying forward useful context, following preferences and constraints, and staying current over time. The company gave examples such as remembering a user’s camera setup for product recommendations, or using a stated dietary preference in future planning.
According to OpenAI, the new architecture is designed to make those kinds of responses more accurate and relevant across many conversations and longer time periods. The company said the rollout starts now for Plus and Pro users in the US, with broader availability to follow.
