FAQ

Quick answers about ChatGPT's dreaming memory.

What is dreaming?
A background process (first introduced April 2025) that automatically curates ChatGPT's memory by referencing your chat history — synthesizing fresh, relevant context without you saying “remember this.” The June 2026 release is its most capable version, Dreaming V3.
When did it launch and who gets it?
June 4 2026 for Plus and Pro users in the US, rolling out to additional countries and to Free and Go users over the coming weeks.
What does “good memory” mean to OpenAI?
Three things: carry forward useful context, follow your preferences and constraints, and stay current as time passes. It evaluates each across 2024 (saved memories), 2025 (+ Dreaming V0) and 2026 (Dreaming V3).
How does it remember me?
A background process learns from many conversations at once and synthesizes a memory state, rather than only writing notes during a single chat. You can review the result in the memory summary page.
Are there benchmark numbers?
OpenAI reports improvements as charts and doesn't publish citable percentage figures in the announcement, so this site doesn't quote any. The one concrete number given is a ~5× reduction in the compute needed to serve dreaming to Free users.
How do I see or turn off what it remembers?
Use the memory summary page to view, add, update and set topic instructions; manage or clear memory in Settings → Personalization → Memory; or use a Temporary Chat that doesn't use or update memory. See OpenAI's Memory FAQ for exact steps.
What about privacy?
Memory persists separately from individual chats, so deleting one conversation may not erase what was learned. Treat the Privacy & risks page as general context — for exact policy, rely on OpenAI's Memory FAQ and your regional settings.
Is this website official?
No. It's an unofficial bilingual digest of OpenAI's announcement, not affiliated with OpenAI. For anything authoritative, see the original article and the Memory FAQ.