ChatGPT Learns to Dream: Meet the New Memory Called Dreaming!

Hey there, it's me, Shiichan! Today I found an update with a wonderfully dreamy name. What does it mean for ChatGPT to "dream"? Let's dig in together.

OpenAI News openai.com

What was announced?

From OpenAI's News comes a big rebuild of ChatGPT's memory, called "Dreaming." The post is titled "Dreaming: Better memory for a more helpful ChatGPT," and it went live on June 4, 2026.

Instead of the old "saved memories" list, where you had to ask ChatGPT to remember specific things, Dreaming works in the background, reading back through your past conversations and updating what ChatGPT remembers all on its own. Just like the name suggests, it's as if ChatGPT reshapes its memories while you're away, like dreaming.

Why it matters

The old memory was handy, but a little clumsy. One issue was that memories went stale over time. If it remembered "you're going to Singapore in July," it might keep saying "going to" long after the trip was over.

The other issue was that it only remembered things you explicitly asked it to. Important context that came up naturally in conversation would slip away. Dreaming tackles both of these at once, so memory gets a lot smarter.

What changes

The biggest change is that memory now understands time. In the Singapore example, once the trip is over it rewrites the memory into something like "you went to Singapore in July 2026." Pretty neat, right?

It also synthesizes across many conversations in the background, lining up the freshest and most relevant context every time. That means fewer "wait, don't you remember what I told you?" moments, and smoother chats overall. And because it captures context that shows up naturally, you don't have to say "remember this" every time, so there's less work for you.

Dive Deep

The rollout starts with Plus and Pro users in the United States. Free, Go, and international users are expected to follow over the coming weeks, so hang tight if you're waiting.

The evaluation numbers are fun too: on OpenAI's internal test of how accurately it recalls facts, success went from 41.5% with 2024's saved memories, to 67.9% in 2025, and up to 82.8% with Dreaming. It's fascinating that just changing how memories are formed moves recall accuracy this much.

Exact settings and finer details may shift depending on the rollout, so peek at your settings once it reaches you.

Wrap-up

• OpenAI announced "Dreaming," a new memory approach for ChatGPT
• It reads back over past conversations in the background and updates memory automatically
• It fixes the "memories go stale" and "it only remembers what you ask" problems
• Memory now understands time, rewriting "planning a trip" into "took the trip"
• On internal evals, factual recall climbed from 41.5% -> 67.9% -> 82.8%
• Plus and Pro in the US first, with Free, Go, and international to follow in a few weeks

If you use ChatGPT a lot and wish it understood you a bit better, this one's for you.