OpenAI announced today a major upgrade to ChatGPT’s memory system, introducing a new architecture that expands the “dreaming” process and promises more personalized, up‑to‑date responses. The change goes live for Plus and Pro subscribers in the United States and will soon be available to free‑tier users.
The original memory feature, launched in April 2024, relied on “saved memories” that required explicit prompts from users. Those memories often grew stale and offered limited context. Over the past year OpenAI layered a background “dreaming” engine that silently stitched together information from multiple chats, but the company admitted it fell short of a full‑scale memory solution.
The new design builds on dreaming by generating a “memory summary” that users can view, edit, or delete at any time. The summary aggregates key facts about a user’s preferences, past conversations, and any self‑reported details. When a user asks a question, ChatGPT can pull from that summary automatically, reducing the need for repetitive instructions.
OpenAI pairs the summary with a “sources” feature introduced alongside GPT‑5.5 Instant. Sources let users see exactly which pieces of stored information the model used to craft a response. By exposing the underlying data, the company hopes to give users more transparency and control over how their personal context influences answers.
In practice, the upgrade could change everyday interactions. A photographer who once told the bot about a specific camera model will see product recommendations that match that gear without re‑mentioning it. A traveler who discussed a Singapore itinerary will receive suggestions that incorporate preferred street‑photography spots, because the system remembers the earlier conversation.
OpenAI says the revamped architecture is also more compute‑efficient, allowing the dreaming process to run at scale for free‑tier accounts. While Plus and Pro users will notice increased memory capacity immediately, free users can expect the first “dreaming‑based” memories to appear in the coming weeks as the rollout expands internationally.
The company plans to extend the feature to additional countries over the next several weeks. By automating memory updates and pruning outdated references, OpenAI aims to avoid situations where the bot mistakenly treats a past trip as an upcoming event. The move marks the most significant step yet toward a continuously learning, personalized AI assistant.
Cet article a été rédigé avec l'assistance de l'IA.
News Factory SEO vous aide à automatiser le contenu d'actualités pour votre site.