ChatGPT Dreaming: How the New Memory System Works

ChatGPT Dreaming: How the New Memory System Works

ChatGPT's Dreaming V3 automatically learns who you are from past conversations. Here's what changed, what's free, and what it means for your privacy.

Too much jargon?→ Look it up in the glossary

Imagine you've been chatting with ChatGPT for months — about travel, work, random projects. Then you ask for a recommendation, and ChatGPT says: "Right, you mentioned you don't like long flights." Without you bringing it up again. Just like that.

That's new. That's "Dreaming V3". On June 4, 2026, OpenAI completely rebuilt ChatGPT's memory system.

What was there before?

ChatGPT used to have a simple notebook. You could manually tell it things to remember: "I'm a vegetarian", "I work as a teacher." The AI would pull those notes when relevant. Useful — but static. And most users never really engaged with it.

What does Dreaming V3 do differently?

Dreaming V3 works automatically. A background process reads through your old conversations and builds a picture of you — without any input from you. Three principles govern what gets stored:

Freshness: Recent information wins over outdated data. You took a vacation last summer? ChatGPT knows that's past travel, not planned.

Continuity: Conversations feel connected even when weeks have passed between them. You don't have to re-introduce yourself every time.

Relevance: Not every offhand remark ends up in long-term memory. The system filters out noise.

The result: ChatGPT genuinely gets to know you over time. Not like a stalker — like a good assistant who listens and remembers what matters.

What's free?

More than you'd expect. OpenAI is rolling out Dreaming V3 to free users too — as of now the rollout is in progress, starting with paid subscribers (Plus, Pro). Paid users get more storage, deeper analysis, and better recall of older conversations. For everyday use, the free version will work well once it arrives. OpenAI's exact feature breakdown, as usual, trickles out gradually.

What you can try right now: check what ChatGPT has already stored about you. Settings → Personalization → Manage memories. This works even without Dreaming V3 — and it's a useful first look.

Privacy: what you should know

ChatGPT now systematically reads through your old conversations. That's new. Worth a moment's thought.

OpenAI gives you control: you can view, edit, and delete all stored memories under Settings → Personalization. If you don't want automatic analysis at all, use Temporary Chat — it runs with no memory whatsoever.

Dreaming only analyzes your own conversations, not anyone else's. What the system "knows" about you is visible to you. Compared to many other services, that's relatively transparent — though trust ultimately remains trust.

Is this for you — or just tech enthusiasts?

For regular users, this is genuinely useful. Anyone who uses ChatGPT regularly for similar things — writing, research, planning — will notice that responses feel better calibrated over time. No setup, no explaining yourself. The system learns from the conversation.

And if you're curious: just check what ChatGPT has stored about you. Settings → Personalization → Manage memories.

Sometimes surprisingly accurate. Sometimes completely off.

That's the current state of AI: learning, but not infallible. Dreaming V3 is a real step forward — and a good moment to check what an AI thinks it knows about who you are.