ChatGPT Gets Smarter Memory as OpenAI Introduces ‘Dreaming’ System

98
06 Jun 2026
min read

News Synopsis

OpenAI has rolled out a significant upgrade to ChatGPT, introducing a new memory architecture designed to make conversations more personalized and context-aware. The update aims to help the AI assistant better remember user preferences and past interactions while improving efficiency and accuracy.

A Major Leap in ChatGPT’s Memory Capabilities

OpenAI has unveiled an advanced update to ChatGPT’s memory system, marking a major step toward more intelligent and personalized AI interactions. The new feature enables the chatbot to retain and use information from previous conversations more effectively, reducing the need for users to repeat details over time.

This enhancement is part of OpenAI’s broader effort to make AI systems more context-aware and useful for long-term engagement. By improving memory handling, ChatGPT can now better support ongoing projects, preferences, and personalized interactions.

Introducing the ‘Dreaming’ Memory Architecture

At the core of this upgrade is a redesigned system known as “Dreaming.” According to OpenAI, this architecture operates behind the scenes to organise, refine, and update stored information gathered from earlier conversations.

Unlike earlier versions, which relied on static or user-directed memory storage, Dreaming actively manages data to ensure that only relevant and up-to-date information is retained. This allows ChatGPT to maintain continuity across interactions while avoiding the accumulation of outdated or unnecessary details.

The system has been engineered to function efficiently at scale, enabling it to support a growing global user base without compromising performance.

Gradual Rollout to Users

OpenAI has begun rolling out the upgraded memory feature to ChatGPT Plus and Pro subscribers in the United States. The company has indicated that support for additional regions, as well as users on Free and Go plans, will be introduced in the coming weeks.

This phased rollout reflects OpenAI’s strategy of testing and refining new capabilities before making them widely available.

Building on Previous Memory Features

The latest update builds on earlier developments in ChatGPT’s memory functionality. In 2024, OpenAI introduced “Saved Memories,” allowing users to explicitly instruct the chatbot to remember certain details.

In the following year, the company expanded these capabilities by enabling ChatGPT to reference past conversations using an earlier version of the Dreaming system.

The current upgrade represents a more advanced and automated approach, where the AI itself plays a greater role in deciding what information to retain and how to use it effectively.

Focus Areas: Context, Preferences, and Relevance

OpenAI has highlighted three primary areas of improvement in the new memory system:

1. Retaining Useful Context

ChatGPT can now better recall important details from previous interactions, helping maintain continuity in long-term conversations.

2. Applying User Preferences

The system is more consistent in recognizing and applying user preferences, such as writing style, tone, and recurring interests.

3. Maintaining Relevance

Dreaming actively updates stored information to ensure it remains accurate and meaningful as user needs evolve over time.

Together, these improvements aim to create a more natural and intuitive user experience.

Performance Gains Backed by Internal Benchmarks

OpenAI has shared internal benchmark data demonstrating significant improvements in the new system’s performance.

  • Factual recall accuracy increased from 67.9% to 82.8%
  • Preference adherence improved from 55.3% to 71.3%
  • Time-sensitive memory performance rose from 52.2% to 75.1%

These gains indicate that ChatGPT is now better equipped to remember facts, adapt to user preferences, and handle dynamic information over time.

New Memory Summary Page for Users

As part of the update, users will gain access to a dedicated memory summary page. This feature provides transparency into what ChatGPT has learned and stored from past interactions.

The page allows users to:

  • Review stored information
  • Edit or correct details
  • Delete unwanted data
  • Specify new preferences or information to remember

This level of control ensures that users remain in charge of their data while benefiting from enhanced personalization.

Improved Efficiency and Scalability

Another key aspect of the Dreaming architecture is its improved efficiency. OpenAI reports that the new system requires significantly fewer computing resources compared to earlier implementations.

These optimizations make it possible to:

  • Expand memory capacity for existing users
  • Deliver the feature to a broader audience
  • Maintain performance even as usage grows

By reducing resource demands, OpenAI aims to scale the feature globally while keeping costs manageable.

Implications for the Future of AI Interaction

The introduction of a more advanced memory system represents an important milestone in the evolution of conversational AI. As systems like ChatGPT become better at remembering and adapting, they are likely to play a larger role in both personal and professional contexts.

From managing ongoing projects to offering tailored recommendations, enhanced memory capabilities could transform how users interact with AI on a daily basis.

Conclusion: Toward More Human-Like Conversations

OpenAI’s latest upgrade brings ChatGPT closer to delivering truly personalized and context-aware interactions. By combining improved recall, smarter preference handling, and efficient resource use, the Dreaming architecture sets a new standard for AI memory systems.

As the feature continues to roll out globally, it is expected to redefine user expectations for conversational AI, making interactions more seamless, intuitive, and meaningful over time.

Podcast

TWN Exclusive