ChatGPT now remembers and keeps memory fresh over time
OpenAI just overhauled how ChatGPT builds and maintains memory. The new system, called Dreaming V3, moves away from static saved notes toward a background process that continuously synthesizes what ChatGPT knows about you based on your actual conversations.
With this memory model, ChatGPT can carry forward useful context, applies your preferences without being reminded, and updates itself when your situation changes.
Here are the key updates:
- Background memory synthesis: Instead of only saving what you explicitly ask it to remember, ChatGPT now reviews your conversation history in the background and builds a living picture of your context, projects, constraints, and preferences.
- Preference-aware responses: ChatGPT applies what it knows about you automatically. If you are vegetarian, planning travel, or working within specific constraints, it factors those in without you needing to repeat yourself.
- Time-aware memory: Old memory systems went stale. If ChatGPT knew you were traveling to Singapore in July, it would keep acting like you were still there in August. Dreaming V3 updates as time passes, so responses stay relevant to where you actually are now.
- Reviewable memory summary: Users can see a plain-language overview of what ChatGPT knows about them, make corrections, and control what topics it should or should not raise.
- 5x compute reduction: OpenAI reduced the compute cost of this system by approximately 5x.
The new memory system is available now for Plus and Pro users in the US. They are rolling out to additional countries and Free users over the coming weeks.
For busy leaders, better memory can make recurring work much more useful: weekly planning, strategy drafts, board prep, team updates, coaching, and decision support.
But it also raises a practical leadership question: as AI starts remembering more context, teams will need clearer habits around what should be shared, what should be excluded, and how sensitive information is handled.
Read the full announcement from OpenAI.
Read more AI news after this:
