OpenAI rolls out a more scalable ChatGPT memory system

OpenAI is making memory a larger part of the ChatGPT product rather than a small saved-notes feature. In a June 4 research and product post, the company said it is rolling out a more capable Dreaming-based memory system for Plus and Pro users in the U.S., with wider availability planned for additional countries and free users over the coming weeks. The update is meant to synthesize useful context from many conversations, make that context reviewable through a memory-summary page, and handle time-sensitive facts that otherwise go stale. OpenAI says recent improvements cut the compute required to serve Dreaming to Free users by roughly 5x. The important product signal is that personalization is becoming infrastructure: the model is not only answering a prompt, it is maintaining a user context layer that must be fresh, editable, auditable, and economical at hundreds-of-millions-user scale.

Key details: June 4, 2026, OpenAI, ChatGPT memory, Dreaming V3, Plus and Pro users in the U.S., Free rollout planned, About 5x compute reduction claimed.

Continue swiping for more AI Brief stories.