OpenAI is rolling out major improvements to ChatGPT’s memory capabilities, with free users getting access to memory features for the first time. The company announced a new memory architecture that builds on its “dreaming” technology to make the AI chatbot better at remembering and using information from past conversations.
The upgrade addresses longstanding issues with ChatGPT’s memory system. Since OpenAI first introduced saved memories in April 2024, users complained that the feature was too basic and required explicit prompts to work properly. Memories also became stale over time, losing their relevance in new conversations.
This represents a significant shift in how AI chatbots handle personalization. While competitors like Google’s Bard and Anthropic’s Claude have focused on improving their core reasoning abilities, OpenAI is betting that better memory will be the key differentiator. The move also signals OpenAI’s strategy to hook free users with compelling features before converting them to paid plans.
The new system works differently from the old approach. Instead of waiting for users to explicitly tell ChatGPT to remember something, the dreaming process runs continuously in the background. It analyzes conversations and automatically creates what OpenAI calls “memory summaries” that users can read and edit at any time.
Here’s what users can expect from the improved memory system:
- Automatic synthesis of information across multiple conversations
- Readable memory summaries that show what ChatGPT has learned about you
- Ability to add, update, or delete personal information and preferences
- Better context carryover between different chat sessions
- Automatic memory updates as circumstances change over time
The practical benefits become clear in everyday use cases. If you’ve previously discussed your camera gear with ChatGPT, it will remember your specific model when you ask for lens recommendations months later. For travel planning, the AI might suggest photography spots in Singapore if it knows you’re interested in street photography based on past conversations.
OpenAI says the new architecture is also more compute-efficient, which explains how the company can now offer memory features to free users. Previously, the processing power required for the dreaming system limited it to paying customers only.
The memory improvements work alongside OpenAI’s recently released memory sources feature, which came with GPT-5.5 Instant. This feature shows users exactly which remembered information ChatGPT used to personalize a response, giving people more control over their AI interactions.
Plus and Pro subscribers in the United States will get access to the enhanced memory system starting today, with international users following in the coming weeks. Free users won’t get the full memory summary interface, but they will benefit from the dreaming process recording their preferences and information for the first time.
The timing of this release is notable as AI companies race to make their chatbots more useful for everyday tasks. Memory capabilities could be crucial for AI assistants to move beyond one-off conversations and become genuinely helpful digital companions. OpenAI’s decision to extend memory features to free users also reflects the competitive pressure in the AI space, where companies are using premium features as acquisition tools.
For existing ChatGPT users, the changes should be mostly seamless. The AI will start building memories from new conversations while gradually improving its understanding of long-term users based on their chat history. Users who prefer not to have their conversations remembered can still disable the memory feature entirely in their account settings.
