OpenAI is rebuilding how ChatGPT remembers you. The company says it has begun rolling out a more capable memory system built on a background process it calls 'dreaming,' which quietly synthesizes what ChatGPT knows about a user across many conversations to keep that context fresh and relevant.
The update reached Plus and Pro users in the US on Thursday, OpenAI says, with free and lower-tier users and more countries to follow over the coming weeks.
From notes to 'dreaming'
Memory started simply. OpenAI launched 'saved memories' in April 2024 — ChatGPT wrote things down when you asked it to, like 'remember I'm traveling to Singapore in July.' The company concedes those notes went stale and missed anything you didn't explicitly flag.
In April 2025 it added the first version of dreaming: a background process that curates memory by referencing your chat history rather than waiting for an instruction. The version shipping now, which OpenAI calls Dreaming V3, makes that the main memory system rather than a supplement to saved notes.
Keeping memory current
The pitch is that memory should age. OpenAI gives the example of a trip: 'You're going to Singapore in July' should become 'You went to Singapore in July 2026' once the date passes, so recommendations snap back to your home city. The synthesized memory is reviewable on a summary page, where users can add, edit, or tell ChatGPT what to bring up and when.
Why it matters
The headline change is reach. OpenAI says it cut the compute needed to serve dreaming to free users by about five times, which is what lets it extend the system beyond paying tiers and raise memory limits for Plus and Pro. A model that remembers more, for more people, is more useful — and a deeper store of personal context, which is why the company points users to its memory controls and FAQ.