OpenAI launches the new generation ChatGPT memory system Dreaming to optimize freshness, continuity, and relevance

BlockBeats News, June 5th, OpenAI announced the launch of a more powerful and scalable ChatGPT memory synthesis system called Dreaming, designed to enhance the freshness, continuity, and relevance of memory, and to address issues of obsolescence, accuracy, and scalability when applying memory across hundreds of millions of users and over many years. OpenAI stated that the memory feature can help ChatGPT learn user preferences, projects, and limitations, enabling future conversations to be based on shared context rather than starting from scratch. This update is now available to US Plus and Pro users today, and will be rolled out to more countries and Free and Go users in the coming weeks.

According to the introduction, ChatGPT first launched the "Save Memory" feature in April 2024, allowing users to ask ChatGPT to remember specific information. In April 2025, OpenAI introduced the first version of Dreaming, enabling ChatGPT to reference chat history through background processes, automatically organize, and synthesize memory states. The newly released Dreaming V3 is a new generation memory architecture built on this mechanism.

OpenAI states that the new system has improvements in factual recall, preference adherence, and maintaining accuracy over time. Its evaluation shows that success rates for factual recall tasks increased from 41.5% in 2024 to 82.8% in 2026; preference adherence success rates increased from 31.4% to 71.3%; and the success rate for maintaining correctness over time increased from 9.4% to 75.1%. OpenAI indicated that recent improvements have reduced the computational power required to provide Dreaming to Free users by about five times, enabling the future rollout to Free users and increasing memory capacity for Plus and Pro users in the coming weeks.