OpenAI is introducing a much more powerful and computing-efficient memory system for ChatGPT that will tackle the problems of staleness, correctness, and scalability for the massive global user base. This major update from the “dreaming” system that was launched for ChatGPT in 2025 and that curated memories automatically by referencing chat history in the background is a huge departure from the “Saved memories” that was the first feature released in April 2024. The newly synthesized memory states that the new architecture is using is providing relevant information all the time, it is, in fact, learning user preferences, projects, and constraints without requiring explicit instructions.
Also Read: OpenAI Unveils GPT-5.5, Advancing Agentic AI for Real-World Workflows
It will be able to keep the useful context, follow user-imposed constraints, and stay relevant to the changing situations. This system will be capable of remembering useful context, following specific user limitations, and even staying up to date as the situations change, like keeping track of project deadlines or new personal requirements. Through a special summary page, users will be able to manage these synthesized memories by reading highlights, changing details or even giving new instructions, Because of this ensuring a personalized experience that keeps changing with the users. OpenAI emphasizes that “Memory is what helps ChatGPT learn your preferences, projects, and constraints, allowing future conversations to start from shared context rather than from scratch.” Currently available to Plus and Pro users in the U.S., the update will reach additional countries, Free users, and Go users in the coming weeks, promising a more intuitive and context-aware interaction model that drastically improves factual recall and preference adherence.
