OpenAI on Thursday announced the rollout of Dreaming V3, a rebuilt memory architecture for ChatGPT that automatically synthesizes and updates user context in the background, replacing the older approach that relied on manually saved memories and often became stale over time. This development represents a meaningful evolution in how AI assistants maintain and apply knowledge about their users, signaling OpenAI’s growing emphasis on personalization as a central pillar of the ChatGPT experience. By moving away from a system dependent on manual upkeep, the company aims to deliver a more seamless and responsive assistant that adapts naturally to the changing circumstances of those who rely on it.
The new system is available immediately to Plus and Pro subscribers in the United States, with expansion to Free and Go tier users planned over the coming weeks. This phased rollout reflects a deliberate strategy of introducing the capability first to paying subscribers before extending it more broadly, allowing the company to refine the experience while gradually scaling access to its vast user base. The planned expansion to free-tier users in particular signals OpenAI’s intention to make sophisticated memory capabilities a standard feature rather than a premium add-on reserved for paying customers.
Dreaming V3 represents the third iteration of a feature that OpenAI first introduced in April 2025, when ChatGPT gained the ability to reference past chat history beyond its explicit saved-memories list. The original “Dreaming” concept allowed background processing to organize conversational context, laying the groundwork for a more intelligent and autonomous approach to memory. The new version builds substantially upon this foundation, employing a more computationally efficient architecture capable of serving as ChatGPT’s primary standalone memory system. This architectural advance marks a notable maturation of the feature, transforming it from an experimental supplement into a robust and central component of the platform.
The mechanics of how Dreaming V3 manages user context reveal a thoughtful approach to maintaining relevant and accurate information. According to OpenAI’s announcement, the system evaluates memories against three distinct criteria: carrying forward useful context from prior conversations, following user preferences and constraints, and keeping information current as circumstances change. To illustrate this dynamic capability, the company offered the example of a user returning home from a trip to Singapore, after which the system updates its understanding so that it no longer recommends restaurants in that city at odd hours, instead reverting to appropriate local suggestions. This example underscores the system’s capacity to recognize and adapt to shifts in a user’s situation, a quality that distinguishes it from more static approaches to memory.
Importantly, the system preserves user agency and control over the information it retains. Users can view, add, or edit what ChatGPT knows about them through a dedicated memory summary page, and they can set topic-specific triggers that determine when certain context should be surfaced. This emphasis on transparency and user control addresses an important consideration in the design of memory systems, ensuring that individuals retain meaningful oversight of the information the assistant holds and how it is applied. Such features reflect an awareness of the trust required when an AI system maintains a persistent understanding of its users over time.
A key enabler of the wider rollout is a roughly fivefold reduction in the compute cost of running the memory system for free-tier users, according to OpenAI. This substantial efficiency gain paves the way for extending the feature beyond paying subscribers to the platform’s broader user base, which exceeds 200 million monthly active users. The ability to deliver advanced memory capabilities at significantly lower computational cost represents an important technical achievement, as it makes the economics of offering such features to a massive free-tier audience considerably more viable. This efficiency improvement therefore plays a crucial role in OpenAI’s ambition to democratize access to personalized AI experiences.
The Dreaming V3 launch arrives amid a broader effort by OpenAI to build out ChatGPT as a comprehensive personalized assistant. In May, the company launched a personal finance experience for Pro users that connects bank accounts via Plaid, and earlier that same month it shipped memory improvements for Plus and Pro tiers that enhanced contextual retrieval from past conversations, saved memories, and connected files. These successive developments illustrate a clear and consistent trajectory, with OpenAI steadily expanding the range of ways in which ChatGPT can understand and serve the individual needs of its users across diverse aspects of their lives.
Ultimately, the Dreaming V3 launch marks a significant shift in how OpenAI conceives of memory, moving it from an auxiliary feature to what the company describes as a core capability of ChatGPT. This repositioning sets the stage for deeper personalization and reflects the intensifying competition among leading AI developers, as OpenAI vies with Anthropic and Google in the race to build AI assistants capable of maintaining meaningful, long-term relationships with their users. As these companies continue to innovate in the realm of memory and personalization, the nature of human interaction with AI assistants appears poised to grow richer and more individualized. The introduction of Dreaming V3 stands as a notable milestone in this ongoing evolution, offering a glimpse into a future in which AI assistants understand their users with increasing nuance and adapt fluidly to the ever-changing contours of their lives.
