OpenAI rolled out a revamped memory architecture for ChatGPT, extending the “dreaming” system that stitches together user preferences and past interactions. The upgrade promises faster, more accurate recall across multi‑year horizons, addressing earlier issues of staleness and scalability, and improves response relevance for complex queries. Plus and Pro subscribers in the United States receive the new engine for their accounts today, with doubled storage capacity.

OpenAI also announced that the same dreaming‑based memory will debut for free users after a five‑fold drop in compute demand. The efficiency gain lets the company meet its quality bar while serving the feature at scale. Free accounts will automatically inherit the enhanced context handling, though users can revert to the legacy saved‑memories mode in settings, across all supported languages and platforms globally.

By unifying memory across paid and free tiers, OpenAI narrows the functional gap that has long separated casual users from subscribers. Developers can now rely on more persistent context when building plugins, and consumers will see fewer repetitive prompts. The rollout marks the most significant expansion of personalized AI assistance since the feature’s initial launch.