ChatGPT Dreaming V3: AI Memory Gets Smarter -- And Now Available to Free Users

Sound familiar? You tell ChatGPT for the third time that you now live in Munich -- not Berlin anymore. Or it keeps suggesting Python even though you switched to TypeScript months ago. ChatGPT's old memory was like a notebook that never got cleaned up: new notes were added, old ones stayed even when they were long outdated.

What is Dreaming V3?

Dreaming is a background process that regularly reviews, verifies, and re-synthesizes all stored memories -- similar to how our brain processes experiences during sleep. Instead of hoarding individual notes, ChatGPT now creates a coherent, up-to-date summary of what it knows about you.

Key improvements:

1. Time-aware memory
This is the most exciting change: ChatGPT now understands that things change over time. If you say in May 'I am going to Singapore in July,' it automatically becomes 'You went to Singapore in July 2026' once the trip has passed. Outdated plans, completed projects, and changed preferences are automatically updated instead of lingering as zombie memories.

2. Significantly better recall accuracy
OpenAI shares concrete numbers from internal evaluations:
- Factual recall: from 67.9% to 82.8% correct memories
- Preference adherence: from 55.3% to 71.3% -- ChatGPT remembers better how you work, what you like, and what you don't
- Accuracy over time: from 52.2% to 75.1% -- outdated information is detected and corrected more reliably

3. Cost reduction brings memory to the free tier
A roughly 5x reduction in compute costs allows OpenAI to offer the full Dreaming system to free-tier users for the first time. Previously, full memory was reserved for Plus and Pro subscribers. The rollout for Free, Go, and international users is coming in the following weeks.

4. Clear memory overview page
Under Settings > Personalization > Memory, you now find a summary page showing what ChatGPT has stored about you, including when the last update occurred.

How to make the most of Dreaming:

- Memory check: Regularly visit the memory page and verify that stored information is accurate. Delete incorrect or unwanted entries individually.
- Actively correct: Tell ChatGPT directly when something has changed: 'I no longer work at Company X, I am now at Company Y.' Dreaming will process this and update the context.
- Temporary Chat: When discussing something you don't want ChatGPT to remember, use Temporary Chat mode. This mode creates no memories and does not access existing ones.
- Pro tip: At the start of a new conversation, say: 'What do you know about me?' This immediately shows you what context ChatGPT is working with, so you can correct any errors.

Privacy note:

Dreaming analyzes your entire chat history in the background to synthesize memories. Key points:

- Opt-out applies: If you have disabled data use for model training in your settings, this also covers Dreaming memories.
- Deletion is multi-step: 'Forget this' or 'Don't mention this again' is not enough. To truly remove information, you must delete it from all sources: the memory page, past chats, archived chats, uploaded files, and connected apps.
- Full disable: You can turn off memory entirely under Settings > Personalization > Memory and delete all data.
- Sensitive data: Never share personal information like tax IDs, passwords, health data, or bank details with ChatGPT -- even in a chat that's 'just quick.' Dreaming could incorporate this information into your summary.

Who benefits most?

Anyone who uses ChatGPT regularly -- especially for recurring tasks like writing emails, editing text, programming, or brainstorming. The more context ChatGPT has about your work style, projects, and preferences, the better the responses become. Dreaming V3 ensures this context stays current instead of going stale.

Sources: openai.com/index/chatgpt-memory-dreaming, help.openai.com/en/articles/8590148-memory-faq, engadget.com/chatgpt-s-memory-is-getting-better-especially-if-you-re-on-the-free-tier