OpenAI ChatGPT Dreaming — Better Memory for a More Helpful AI

What Is "Dreaming" and Why Does It Matter

On June 4, 2026, OpenAI began rolling out a new memory system for ChatGPT called "Dreaming" — a fundamentally improved architecture designed to tackle the staleness, correctness, and scalability challenges that memory systems face when applied to hundreds of millions of users across multi-year time horizons.

Memory is what helps ChatGPT learn your preferences, projects, and constraints, allowing future conversations to start from shared context rather than from scratch. Over the past two years, memory has grown into a critical part of the ChatGPT experience, helping the AI better understand your context so it can help you accomplish meaningful goals over time. This update is available to Plus and Pro users in the US today, with plans to roll out to additional countries and Free users over the coming weeks.

The Evolution of Memory — From 2024 to 2026

Memory first launched in April 2024 as "Saved Memories." This feature let you ask ChatGPT to remember information and carry it forward into future chats. However, this system was like talking to someone who took a few notes but forgot everything that wasn't written down. Saved memories also tended to go stale over time, eventually becoming incorrect or irrelevant.

In April 2025, OpenAI updated ChatGPT's memory by giving the model the ability to reference chat context outside the saved memories list — introducing the first version of Dreaming V0, a method for ChatGPT to automatically curate memories in the background by referencing chat history. Unlike saved memories, dreaming used a background process that allowed ChatGPT to learn from many conversations and synthesize its memory state — always providing the freshest, most relevant context.

Today, in 2026, OpenAI is launching Dreaming V3 — a significantly more capable and compute-efficient memory architecture built on top of dreaming. This is a system that automatically updates memories as time passes, maintaining relevance across months and years of interaction.

Three Criteria — How "Good Memory" Is Evaluated

OpenAI evaluates memory quality against three core criteria:

• Carry forward useful context: You tell ChatGPT something once, and it remembers in subsequent chats. For example, if you discussed your underwater photography equipment, ChatGPT should know your camera model and give compatible recommendations.
• Follow preferences and constraints: If you're vegetarian, ChatGPT should always offer vegetarian options. If you prefer quiet dinners over crowded bars — that should be factored in too.
• Stay current over time: Memory should account for the passage of time. "The user is planning a trip to Singapore in July" — when the trip ends, memory should update: "The user went to Singapore in July 2026."

Based on these three criteria, OpenAI compares memory systems across three eras: 2024 (Saved Memories), 2025 (Saved Memories + Dreaming V0), and 2026 (Dreaming V3). The results clearly show that Dreaming V3 significantly outperforms previous versions on all three criteria.

Practical Examples — How the Difference Plays Out

The examples OpenAI presented are remarkably illustrative. Imagine you're planning a trip to Singapore and ask ChatGPT for help with your itinerary.

Without memory: ChatGPT gives you a generic Singapore travel guide — standard tourist attractions, generic restaurant recommendations, typical advice. This is what any search engine could provide.

With memory: ChatGPT knows your interests — wildlife photography, quiet dinners, the need for strong air conditioning in hotels. It suggests Bird Paradise + Night Safari for photography, recommends reservation-based restaurants instead of crowded hawker centres, and prioritizes AC when suggesting hotels. This is the personalized experience that only a memory system like Dreaming can deliver.

Another example: if you were in Singapore and then returned home, the old memory system might still think you're in Singapore. Dreaming V3 automatically updates this information — "The user was in Singapore in July, now they're home" — and provides relevant recommendations for your actual location.

The difference between generic and personalized responses isn't marginal — it's transformative. A ChatGPT with good memory doesn't just answer questions; it understands context, anticipates needs, and provides recommendations that feel like they came from a knowledgeable friend, not a search engine.

Technical Progress and Scalability

One of the most significant achievements of the new Dreaming version is computational efficiency. According to OpenAI, recent improvements reduced the compute required to serve dreaming to Free users by approximately 5x. This means OpenAI can now roll out dreaming to Free users and increase memory capacity for Plus and Pro users.

This achievement is technically fascinating. Memory synthesis requires analyzing a user's chat history, extracting context, and presenting it in a relevant format. With hundreds of millions of users, this task becomes exponentially complex. A 5x efficiency improvement means the system is faster, cheaper, and can serve more people.

Users can view their memory summary on the Memory Summary page — seeing what ChatGPT knows about them, with the ability to correct or delete information. This is an important transparency element — users have control over what the AI knows about them. You can also provide instructions on what topics ChatGPT should bring up and when, giving you fine-grained control over the personalization experience.

Implications for Businesses and Developers

OpenAI's memory development is directly relevant to how businesses can use AI tools. If your company uses an AI chatbot for customer communication, the memory system determines how "smart" that chatbot feels to your customers.

Imagine a hotel whose AI assistant remembers that a guest has allergies to certain foods, prefers a quiet room, and last visited three months ago — now returning for a new stay. This is the level of personalization that Dreaming-type systems bring, and it's the kind of experience that turns one-time customers into loyal repeat visitors.

For developers building AI-powered applications, the implications are equally significant. OpenAI is essentially providing a memory infrastructure that you can build on top of. Instead of building your own memory system from scratch, you can leverage ChatGPT's memory capabilities to create deeply personalized experiences for your users.

The competitive landscape is also heating up. Google, Anthropic, and others are working on similar memory systems. The "memory war" has begun, and the company that delivers the best memory experience will have a significant advantage in the AI assistant market. This competition benefits everyone — users get better AI, businesses get more capable tools, and developers get more powerful infrastructure.

What's Next — The Future of AI Memory

Dreaming is just the beginning. OpenAI states this is the "most capable memory system yet" and that improvements will continue. Several directions to watch:

• Free user access: Dreaming will gradually roll out to Free users — making AI personalization accessible to the masses
• Privacy and control: Users have full control over their memory — view, correct, and delete at any time
• Scalability: The 5x efficiency improvement means the system is ready for billions of users
• Competition: Google, Anthropic, and others are building similar systems — the memory war is underway
• Enterprise applications: As memory improves, enterprise AI assistants will become dramatically more useful for customer service, sales, and support

Ultimately, Dreaming represents a step toward AI that truly knows you — not just your last message, but your history, preferences, and life context. This is AI that doesn't just respond, but understands. And for businesses and developers, it's an invitation to build experiences that were previously impossible — where every interaction builds on the last, and every conversation feels like talking to someone who genuinely knows and cares about your needs.

🔗 წყარო: OpenAI