Breaking

OpenAI Reworks ChatGPT Memory With Dreaming System and New Controls

September 27, 2026 at 15:10 EDT

Foundation Models
Sales & Business

OpenAI has reworked ChatGPT memory around a system it calls Dreaming, moving the product toward automatically synthesized, continuously updated user context rather than relying only on explicit saved notes. The change is meant to make ChatGPT more useful for recurring work, personal preferences and long-running projects, while also raising new questions about transparency and user control.

June 4, 2026 | OpenAI

ChatGPT memory becomes a living profile

The new Dreaming system synthesizes context from past chats so ChatGPT can carry preferences, projects and constraints forward while giving users a summary to review.

Lower compute required to serve Dreaming to Free users

Core goals: continuity, preferences and freshness

Days for chat-history-derived memory deletion after opt-out

Compute needed to serve memory

Before

5 relative units

After

1 relative unit

The update made the same class of memory service practical for broader rollout.

How Dreaming changes the loop

Past chats

Context can come from prior conversations and saved memories.

Synthesis

The system updates a broader summary of relevant context.

Personalized answer

Responses can use preferences, projects and constraints.

Why users want it

Less repetition, stronger cross-chat continuity and more tailored help for ongoing work.

Why users worry

Automatic summaries may omit details, compress important context or make control feel less explicit.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.