ChatGPT’s memory gets a major upgrade: here’s how the new ‘dreaming’ system works
For months, users of ChatGPT have complained about inconsistent recall—details vanishing, outdated preferences lingering, and memories that simply wouldn’t stick. Now, OpenAI is rolling out a significant ChatGPT memory upgrade called “dreaming,” designed to make conversations feel more natural and context-aware. This overhaul addresses one of the platform’s most persistent weaknesses since memory features launched in 2024.
How the dreaming system changes memory management
The original memory system relied on explicit cues—users had to ask ChatGPT to remember something. That approach often let valuable context slip through the cracks. Temporary details, like an upcoming vacation or a dietary preference, had no way to update themselves over time.
OpenAI’s new dreaming system works differently. It synthesises information from your chat history automatically, running in the background without waiting for a manual trigger. This means that if you mentioned a trip to Singapore last month, the system no longer assumes you’re still there. It accounts for the passage of time, refreshing or discarding information as context evolves.
In addition, a new memory summary page gives users a clear overview of what ChatGPT knows about them. You can review stored facts, make corrections, and add details manually. This added transparency helps you steer the AI’s memory more precisely.
Better application of preferences in conversations
The update also improves how ChatGPT applies your preferences during chats. If you’ve noted in past conversations that you’re vegetarian or that you prefer quieter restaurants, the new system is better at pulling that context into new discussions. You no longer have to repeat yourself—a common frustration with earlier versions.
This means that the ChatGPT memory upgrade not only stores more accurate information but also uses it more intelligently during interactions. The result is fewer awkward or irrelevant responses, making the chatbot feel more like a helpful assistant that remembers your habits.
Who gets the update and when
OpenAI is rolling out the dreaming system to ChatGPT Plus and Pro users in the United States starting today. Over the coming weeks, it will expand to additional countries and to the Free and Go tiers. The inclusion of the free tier is particularly notable.
According to OpenAI, recent efficiency improvements reduced the compute required to serve the memory feature by roughly 5x. That made a free-tier rollout practical. The company also increased memory capacity by 2x for Plus and Pro users, allowing them to store more context without performance hits.
This wider availability means that more people can benefit from improved recall, regardless of their subscription level. For those who rely on ChatGPT for daily tasks, this is a welcome change.
Will the dreaming system solve user frustrations?
Memory has been one of the most-requested improvements from ChatGPT users. Complaints about inconsistent recall, stale information, and new memories not saving have surfaced regularly on Reddit and other forums. The dreaming system directly targets these pain points by making memory updates automatic and context-sensitive.
However, whether this ChatGPT memory upgrade resolves those frustrations in practice—rather than in benchmark tests—remains to be seen. Early adopters will likely provide feedback on how well the system handles nuanced changes, such as shifting preferences or evolving life circumstances.
For now, OpenAI is positioning dreaming as a significant leap forward. If it works as advertised, users can expect fewer “I don’t remember that” moments and more fluid, personalised conversations. As the rollout expands, the true test will be how seamlessly the system adapts to real-world usage patterns.
To learn more about optimising your ChatGPT experience, check out our guide on improving AI productivity. For updates on future features, visit the OpenAI blog.
