OpenAI is rolling out a major upgrade to ChatGPT"s memory, making the system more capable, current, and scalable across long-term use.
Memory allows ChatGPT to remember useful details about users, including their preferences, projects, and constraints. Instead of starting every conversation from scratch, ChatGPT can use this context to provide more relevant responses in future chats.
OpenAI first launched saved memories in February 2024. That feature allowed users to explicitly ask ChatGPT to save information into its memory, such as travel plans or writing preferences. However, this system had limits because it depended heavily on users giving clear instructions to remember something. Additionally, saved memories could become stale over time.
In April 2025, OpenAI expanded memory by allowing ChatGPT to reference past chat context outside the saved memories list. This was powered by a background process called “dreaming,” which automatically curates memories from chat history. This made ChatGPT better at learning from natural conversation without requiring users to manually save every detail.
Today, OpenAI announced a more capable and compute-efficient memory architecture built on top of dreaming. This new system improves ChatGPT’s ability to carry forward useful context, follow user preferences, and remain accurate as time passes. According to OpenAI’s internal evaluations, the new system improves factual recall from 67.9% in 2025 to 82.8% in 2026. Preference adherence improves from 55.3% to 71.3%, while accuracy over time improves from 52.2% to 75.1%.
The best part of this new system is a new memory summary page where users can review ChatGPT"s memories. Users can even update details, correct information, or give instructions on what topics ChatGPT should bring up and when.
This new, improved memory system is available to ChatGPT Plus and Pro users in the US starting today. It will roll out to more countries, as well as Free and Go users, in the coming weeks.