ChatGPT's Dreaming V3 memory upgrade lets it remember you better across conversations

ChatGPT Memory Feature Gets Major Overhaul with Dreaming V3

OpenAI announced on June 4 that ChatGPT is receiving significant improvements to its memory feature through a new system called Dreaming V31

1

2

2

. The enhanced memory system marks a departure from the chatbot's original note-taking approach, where users had to explicitly instruct ChatGPT to remember specific details. Instead, the new architecture automatically curates information from chat history in the background, allowing the AI to recall context, remember user preferences, and stay current over time without manual prompting4

4

.

The update addresses a persistent frustration among users who had to re-introduce themselves and their preferences with each new conversation3

3

. When you start a new chat with ChatGPT, the AI can now build on prior context, especially for complex, long-running projects. For example, if you've discussed your camera setup in previous conversations, you can simply ask for products compatible with "my photography setup" and receive tailored recommendations1

1

.

How Dreaming V3 Transforms User Experience

The ChatGPT memory system now operates through what OpenAI calls "dreaming"—a process where the AI reviews past chats to identify patterns, preferences, and important information1

1

. This happens in the background while the chatbot isn't actively engaged with users. Dreaming V3 introduces several critical improvements that enhance personalization and user experience.

One standout feature is temporal awareness. The system now understands the passage of time and automatically updates memories accordingly. OpenAI provides an example of a user planning a trip to Singapore in July. Previously, ChatGPT might continue assuming the user was still traveling long after the trip ended. With Dreaming V3, the AI can revise its memory from "You're going to Singapore in July" to "You went to Singapore in July 2026" once the trip concludes2

2

. This temporal understanding prevents outdated context from interfering with current conversations.

Users also gain access to a readable summary of everything ChatGPT knows about them2

2

. You can review this summary, add or update information, and even edit the AI's "dreams" if something has been misremembered1

1

. This transparency matters, especially as OpenAI has indicated plans to use ChatGPT memory for personalized advertising in the future.

Dramatic Performance Improvements and Wider Access

OpenAI evaluated the ChatGPT memory system on three key criteria: carrying forward context, following preferences, and staying current over time4

4

. The results show substantial gains across all metrics. Factual recall task success jumped to 82.8% from 67.9% in 2025 and just 41.5% in 20245

5

. Preference adherence rose to 71.3% from 55.3% in 2025 and 31.4% in 2024. The system's ability to stay current over time reached 75.1%, up from 52.2% in 2025 and just 9.4% in 20245

5

.

Perhaps most significant for AI development is the efficiency breakthrough. Recent improvements reduced the compute requirements to serve dreaming to Free tier users by approximately 5x4

4

5

5

. This makes Dreaming V3 not only more capable but also more practical to deploy at scale. As a result, OpenAI is bringing the dreaming capability to Free tier users for the first time, democratizing access to advanced memory features previously reserved for paying subscribers.

The rollout began with ChatGPT Plus and Pro users in the United States on June 4, with expansion to additional countries and Free and Go users planned for the coming weeks2

2

4

4

. OpenAI is also increasing memory capacity for users on the Plus and Pro plans4

4

.

What This Means for the AI Landscape

The improvements to how ChatGPT can recall context and remember user preferences position it ahead of competitors in the memory domain. While ChatGPT already had one of the better memories among AI agents, rivals like Claude have struggled with memory functionality4

4

. Google has been developing its own memory import tool for Gemini, signaling that contextual memory has become a key battleground in AI development.

The shift toward AI systems that "know you" raises important questions about personalization versus privacy. Users who find the concept unsettling can review exactly what ChatGPT knows, make corrections, or wipe their memory entirely from the new summary page. Those who prefer the old system can also revert to it in settings4

4

.

For users, the immediate benefit is clear: less repetition and more productive conversations. Whether you're managing complex projects, shopping for compatible equipment, or planning travel, ChatGPT can now carry forward the context that matters without constant reminders. As AI assistants become more integrated into daily workflows, memory systems like Dreaming V3 will determine which platforms feel genuinely helpful versus merely functional.