ChatGPT’s memory update arrives—free users get smarter recall

OpenAI is rolling out a new ChatGPT memory architecture built on its earlier “dreaming” work. With it, users get a readable memory summary, more reliable carry-forward of context and preferences, and ongoing memory updates over time. The update starts with Plu

The first time ChatGPT tried to remember you, it mostly needed you to tell it directly.

OpenAI shipped its first memory feature in April 2024, initially known as saved memories. It worked only when users gave strong cues—such as prompting the chatbot to remember a specific fact. People also noticed the memories drifting out of step over time, becoming less relevant as conversations moved on.

So OpenAI spent the following year building a different approach. It began developing what it later called “dreaming,” a process that runs in the background and lets ChatGPT synthesize information from multiple conversations without requiring explicit instructions to store each detail.

“Over the last year, dreaming supplemented saved memories to create a step-function improvement in ChatGPT’s ability to personalize responses and offset the staleness of saved memories,” OpenAI says. “However, it historically was never sufficient as a standalone memory system.”

Today’s release is meant to change that.

OpenAI is now rolling out what it describes as a new memory architecture that builds on dreaming to be both “significantly” more capable and compute-efficient. As ChatGPT synthesizes information about you, it will write a “memory summary” you can read at any time. From there. you can add and update information about yourself. and tell ChatGPT when it should reference what it knows about your preferences.

“If you want to drill down into a particular area to learn more, just chat with the model,” OpenAI says.

The new memory summary is also designed to complement a separate feature OpenAI released alongside GPT-5.5 Instant. That feature—memory sources—lets users see the information ChatGPT used to personalize an answer, and edit or delete that context as desired.

OpenAI argues the update isn’t just about giving users more visibility. It also aims to solve a problem that kept coming up with memory in the first place: relevance over time.

With the new architecture, OpenAI says ChatGPT is better at carrying forward context. For instance. if you’ve talked to ChatGPT about photography before and mentioned the camera you currently use. the chatbot will know to generate tailored product recommendations the next time you ask for suggestions that fit your photography setup.

The same idea applies to preferences. If you’re planning a trip. OpenAI says ChatGPT will use what it has learned from past travel conversations to shape its responses. In the photography example again, it says ChatGPT might suggest a Singapore itinerary that includes spots suited for street photography.

Then there’s the practical part users often worry about but rarely see: whether the bot will keep outdated details around. OpenAI says the system will automatically revise its memories as time passes, so it won’t do something like reference a trip you took in the past as if it were coming up.

The rollout starts with the paying tier. OpenAI is beginning to roll out the new memory architecture to Plus and Pro users in the US starting today. For those users, OpenAI says the improvements translate into greater memory capacity.

Free accounts are next—but not immediately in the same way. Thanks to behind-the-scenes efficiency improvements. OpenAI says ChatGPT will. for the first time. soon start recording memories through the dreaming process for free accounts. The company also says the new architecture will roll out to users in other countries in the coming weeks.

OpenAI ChatGPT memory feature dreaming saved memories memory summary memory sources GPT-5.5 Instant Plus Pro free tier