OpenAI has improved ChatGPT's memory; soon the bot will remember more and this is available for free

OpenAI is improving ChatGPT’s memory and will soon make it available for free

OpenAI announced the launch of a new, more powerful memory system in the ChatGPT chatbot. The new architecture promises “smarter” information storage and will first be available to paid users, then to everyone after a few weeks.

What has changed
Parameter Before Update After Memory volume ~1 GB (approx.) Doubled Precision and relevance Limited, sometimes “forgets” details Improved by Dreaming method Access speed Depends on data volume Reduced thanks to computation optimization

How the new memory works
1. Dreaming Method

- Automatically “sleeps” in the background, filling gaps in the chat history.

- Selects the most relevant messages from past conversations and saves them as memories.

2. Memory Summary

- The user can view a page listing saved facts about themselves.

- Ability to add, edit or delete information.

- Specify topics that ChatGPT should remember and when to mention them.

3. Real‑time Context

- With each new conversation the bot uses accumulated data so it doesn’t start from scratch.

- This makes interaction more natural and personalized.

Key improvement priorities
1. Increase volume – now ChatGPT can store twice as much user information.

2. Reduce computational costs – thanks to Dreaming optimization, resource consumption is cut by about five times, making the feature available even to free users.

3. Scale up – the system can serve hundreds of millions of users without losing memory quality.

How it will affect users
- Paid subscribers (ChatGPT Plus and Pro in the US) will get the updated memory today.

- Free users will be able to use the new feature in the coming weeks, although the available memory volume will be slightly smaller than for paid clients.

Conclusion
OpenAI has taken a step forward in creating a smarter and more scalable memory for ChatGPT. The increased data volume, improved accuracy through Dreaming, and significant cost reduction make this function accessible both to subscribers and, in the near future, to all users.