OpenAI has announced a major upgrade to the memory system of its ChatGPT chatbot. The new architecture, called Dreaming V3, allows the model to more efficiently maintain long-term context, automatically update outdated data, and requires five times less computing resources.
OpenAI has announced a major upgrade to the memory system of its ChatGPT chatbot. The new architecture, called Dreaming V3, allows the model to more efficiently maintain long-term context, automatically update outdated data, and requires five times less computing resources.
This optimization made it possible to make Dreaming V3 available to all users, according to the official OpenAI developer blog .
The update is already available to users of the Plus and Pro paid plans in the US. The technology will begin to roll out to other countries in the coming weeks, as well as open access to users of the Free and Go versions.
The main change is the transition to a new architecture built on the basis of the background process «Dreaming». Unlike the first versions of memory, which required the user to give explicit instructions like «remember this», the new system independently analyzes the correspondence history in the background and highlights the key facts, preferences and limitations of the user.
In addition, the updated Dreaming V3 technology solves the problem of data relevance over time. If earlier artificial intelligence could consider information about a user’s temporary trip to be relevant for a long time, now the system automatically corrects records after the events have ended. For example, the note «user is going to Singapore in July» after the trip is completed is automatically transformed into «user visited Singapore in July 2026».
Users can view all accumulated memories and, if necessary, edit or delete them on a special memory summary page.
«Today, we are launching a significantly more powerful and computationally efficient memory architecture built on the foundation of ‘daydreaming,’» OpenAI notes. It is this 5x computational optimization that has allowed the company to scale the feature to hundreds of millions of users of the free version.
