What is ChatGPT's 'dreaming' memory feature?

Dreaming is OpenAI's new memory architecture that automatically updates stored facts over time. It can revise 'You're going to Singapore in July' to 'You went to Singapore in July 2026' when the trip ends. Memories are auto-summarized, user-editable, and pulled into context across saved memories, past chats, uploaded files, and connected apps like Gmail.

Is dreaming memory free in ChatGPT?

Yes, starting June 4, 2026. OpenAI announced the rollout of the new memory system to free-tier users globally. Free users get the same core 'dreaming' behavior (auto-updating memories, summary view, manual edits) as Plus and Pro, though Plus and Pro retain larger memory budgets and connector access (Gmail, files, etc.).

How do I control or turn off ChatGPT memory?

Open Settings → Personalization → Memory. You can view a readable summary of everything ChatGPT remembers, edit or delete individual facts, turn off memory globally, or use 'Temporary Chat' to bypass memory for a single conversation. As of June 4, 2026, you can also set per-domain reference rules (e.g., 'don't use my health memories during work chats').

How is dreaming different from the older ChatGPT memory?

Older memory was a static list of facts ChatGPT chose to remember. Dreaming continuously rewrites those facts as new information arrives — past tense, expired plans, changed preferences. It also unifies four sources (saved memories, past chats, files, connectors) into one context retrieval, instead of treating them separately.

Quick Answer

ChatGPT Dreaming Memory June 2026: Now on Free Tier

Published: June 5, 2026

ChatGPT Dreaming Memory June 2026: Now on Free Tier

On June 4, 2026, OpenAI rolled out its “dreaming” memory upgrade to free-tier ChatGPT users. It’s the biggest free-tier feature drop since GPT-5.5 Instant became the default in May. Here’s what dreaming actually does, what changed for free users, and how to control it.

Last verified: June 5, 2026

What is “dreaming”?

Dreaming is OpenAI’s new term for ChatGPT’s memory architecture. The key difference from the older memory feature: memories are automatically updated over time instead of being static facts.

The official example from OpenAI: a memory that starts as “You’re going to Singapore in July” gets automatically rewritten to “You went to Singapore in July 2026” once the trip is over. The model essentially “dreams” — consolidating and revising stored context during low-activity periods, much like sleep-stage memory consolidation in humans (though the analogy is more naming than mechanism).

Under the hood, dreaming combines four context sources into one retrieval:

Saved memories — explicit facts ChatGPT has been told to remember
Past chats — recent conversations from your history
Uploaded files — Pro/Plus users with file connectors
Connected apps — Gmail, Google Calendar, etc., where available

When you start a new conversation, ChatGPT pulls the most relevant context from all four sources, not just saved memory.

What changed for free-tier users on June 4

Before June 4, 2026:

Free users had no persistent memory at all
Each conversation was effectively a clean slate

After June 4, 2026:

Free users get the full dreaming memory system
Auto-updating memories
Readable memory summary in Settings → Personalization → Memory
Manual edit/delete controls
Temporary Chat option to bypass memory

Plus and Pro users keep advantages: larger memory budgets, file connectors, Gmail/Calendar integration, and priority on memory retrieval during high-load periods.

Why this matters

ChatGPT free has ~700M weekly active users (last public figure). Adding persistent memory to that base does three things:

Lock-in increases sharply — once ChatGPT knows your preferences, the cost of switching to Claude or Gemini rises significantly
Per-user quality improves — answers become personalized within a few chats
Competitive pressure on Anthropic and Google — Claude’s “dreaming” feature (released May 2026) is currently Pro-only; Gemini’s memory is partially free but more limited

This is OpenAI applying scale as a moat — free tier is the distribution wedge, memory is the retention wedge.

How to control or disable dreaming memory

In Settings → Personalization → Memory:

View memories — readable summary of everything ChatGPT remembers
Edit memories — modify any stored fact
Delete memories — remove individual items or clear all
Turn off memory — disables both reading from and writing to memory
Per-domain rules (new June 4) — e.g., “ignore work memories during personal chats”

For one-off sensitive conversations, use Temporary Chat (the dotted-circle icon next to the model selector). Temporary Chats don’t read from or write to memory.

How dreaming compares to other AI memory systems

Feature	ChatGPT Dreaming (June 2026)	Claude Memory	Gemini Memory
Free tier	✅ Yes (as of June 4)	❌ Pro only	⚠️ Limited
Auto-update memories	✅ Yes (“dreaming”)	✅ Yes	⚠️ Partial
Past-chat recall	✅ Unified	✅ Yes	⚠️ Limited
File/email connectors	✅ Plus/Pro	✅ Pro	✅ Workspace
Per-domain rules	✅ New June 4	❌ No	❌ No
User-editable summary	✅ Yes	✅ Yes	✅ Yes

Privacy considerations

Memory is opt-in but on by default in most regions (EEA and UK users get explicit consent prompts). Memories are tied to your account and not used to train models when memory is enabled (per OpenAI’s current data policy as of June 2026).

If you share an account or use ChatGPT for work, consider:

Use Temporary Chat for sensitive queries
Review the memory summary monthly
Use per-domain rules to compartmentalize work vs personal

Bottom line

Dreaming memory on the free tier is a major free-product upgrade and a strategic move against Claude and Gemini. If you’ve been using free ChatGPT, the experience will get noticeably more personalized over the next few weeks. If you care about control, the new Settings → Personalization → Memory panel is now the most powerful memory dashboard of any major AI assistant.