What the New ChatGPT Memory Feature Actually Is
ChatGPT’s new dreaming-based memory system is an upgraded AI conversation memory feature that quietly collects, updates, and applies your preferences and ongoing context across chats, so you repeat yourself less while keeping strong control over what is stored and how long it stays relevant. OpenAI has turned memory from an explicit note-taking tool into a background layer that can keep track of things like your preferred answer length, long-term projects, dietary rules, or changing travel plans. Instead of relying only on “please remember this” prompts, the dreaming architecture analyzes conversations to decide what is useful enough to keep. At the same time, OpenAI is trying to avoid stale personalization, such as assuming you are still abroad after you return home. The aim is an assistant that feels informed and current, not clingy or outdated.
From Notebook to Dreaming Architecture: What Changed Under the Hood
The original ChatGPT memory from April 2024 worked like a small notebook: you manually saved facts, and the model could reuse them later. Helpful, but narrow. In 2025, OpenAI added an early version of dreaming so ChatGPT could draw from more of your past chat context, not just the saved list. The latest update, often called Dreaming V3, goes further by making this system more capable and efficient enough for wider use. According to Investing.com, factual recall in OpenAI’s internal tests rose from 41.5 percent in 2024 to 67.9 percent in 2025 and now 82.8 percent with the new dreaming memory. Preference-following and freshness also improved significantly, showing that the architecture is better at both remembering and forgetting wisely. Behind the scenes, compute costs to serve dreaming have dropped about five times, which makes scaling this feature more realistic.
How Smarter AI Conversation Memory Changes Day-to-Day Use
For everyday users, the upgraded ChatGPT memory feature mainly means you spend less time repeating instructions. The assistant can remember if you like concise answers, avoid certain topics, or need explanations tailored to a specific skill level. For professionals, the impact is larger. A consultant can keep client context, tone guidelines, and project history alive across weekly sessions. A founder can work on pitch decks, investor lists, and messaging without re-teaching the assistant every time. OpenAI’s own evaluations show preference adherence rising from 31.4 percent in 2024 to 55.3 percent in 2025 and 71.3 percent in the new system, which suggests longer-running work should feel more consistent. As those setup minutes at the start of each session shrink, ChatGPT shifts from one-off answer machine toward an ongoing partner for research, writing, coding, planning, and personal organization.
Control, Transparency, and Staying Current Over Time
More powerful memory raises questions of control and trust, so OpenAI is adding clearer tools around the dreaming architecture. You can review what the system has synthesized through a memory summary page, which highlights key details ChatGPT is using. There, you can correct information, add updates, or guide what should be brought up later. Users can also turn memory off entirely, delete specific memories, reset all memory, or switch to Temporary Chat when they do not want a conversation to read or update stored context. OpenAI distinguishes between saved memories and general chat history, since not everything you say should become long-term context. Internal tests show staying current over time improved from 9.4 percent in 2024 to 52.2 percent in 2025 and 75.1 percent now, reflecting an effort to age out stale details so personalization remains helpful rather than intrusive.
Rollout: Who Gets the New ChatGPT Improvements and When
OpenAI is starting the rollout of the new dreaming-based memory system with ChatGPT Plus and Pro users in one major market, then expanding to more regions and tiers in the coming weeks. The company also plans to extend memory improvements to Free and Go users, supported by the fivefold reduction in compute required to serve dreaming. This staged release helps OpenAI watch how the feature behaves at scale and refine the balance between usefulness and privacy controls. For now, you may notice smoother continuity, better preference-following, and fewer reminders needed when you open new chats. Over time, as memory becomes more widely available and more reliable, ChatGPT edges closer to a working companion that builds an understanding of your projects and style session after session—provided you are comfortable with what it remembers and keep an eye on the settings.
