TL;DR
- Personalization Rollout: OpenAI has released an upgraded ChatGPT memory system to U.S. Plus and Pro users.
- Dreaming Mechanism: The system uses background synthesis to turn past chats into a visible, editable memory summary.
- Free Access: OpenAI says a roughly five-times compute reduction will support Free and Go users in the coming weeks.
- User Control: Editable summaries help users correct stale personal details before they shape later ChatGPT answers.
OpenAI has released an upgraded personalization system on for ChatGPT Plus and Pro users in the United States. The AI assistant there now can retain useful context across conversations.
Free and Go accounts are scheduled to follow during the rollout’s next stage. A background process called dreaming now becomes the main way ChatGPT can synthesize remembered details across conversations.
How Dreaming Changes ChatGPT Memory
Dreaming is OpenAI’s term for a background process that reviews past ChatGPT conversations and turns useful details into context for later answers. Instead of waiting for a user to explicitly ask ChatGPT to remember something, the system can synthesize preferences, projects, and constraints across many conversations, including details that matter only when a later request revives the same task.
Users will see the change through a visible memory summary that ChatGPT can write and revise, while people can read the summary, add to it, update it, and steer when the chatbot should use remembered information. A visible summary gives users one place to inspect what the assistant treats as durable, rather than forcing them to infer those assumptions from later answers.
Users can edit the memory summary when ChatGPT gets details wrong, giving them a way to correct remembered preferences before those details shape a later answer. OpenAI evaluates the system on whether it carries forward useful context, follows user preferences and constraints, and stays current as time passes, such as avoiding a completed trip, expired deadline, or old project state as if it were still upcoming.
Rollout Limits and Assistant-Memory Context
Paid plans get the upgrade first, and Plus and Pro memory capacity is increasing as dreaming becomes the shared foundation for ChatGPT memory. Additional countries, Free plans, and Go users are scheduled for the coming weeks, so free-account access is planned as well. Staged timing separates the current release from the larger test OpenAI still has to run across users who may have different privacy expectations, correction habits, and tolerance for remembered mistakes, especially when remembered preferences influence routine searches, recommendations, or work planning.
Free-user availability raises the same scalability, freshness, and user-control issues, where mistakes can affect casual users who may not monitor settings as closely as paid subscribers. A wider audience also makes the editing path more important: the summary has to be visible enough for users to catch stale preferences before the assistant carries them into another answer. Competitors like Anthropic have also moved toward free-tier memory access, and ChatGPT’s rollout shows durable personalization shifting from a premium convenience toward a baseline assistant feature.
ChatGPT’s memory timeline began with an April 2024 phase that brought ChatGPT memory to Plus subscribers, then added 2025 web-search personalization and 2026 so far memory-source visibility controls.
A separate paid-user push around persistent ChatGPT context also set up the current move toward more durable personalization. Saved memories arrived in April 2024, when ChatGPT depended more heavily on explicit user cues and conversation-time instructions. OpenAI later introduced dreaming as background curation.
