OpenAI is rolling out a major upgrade to ChatGPT’s memory, enhancing the model’s ability to retain context, stay up‑to‑date, and scale across extended interactions.
With this new memory system, ChatGPT can now keep track of user preferences, ongoing projects, and specific constraints, allowing it to deliver more tailored and relevant responses without having to start each conversation from scratch.
In February 2024, OpenAI introduced the first version of “saved memories,” letting users explicitly instruct the model to remember details such as travel itineraries or writing styles. While a useful step, this approach relied heavily on user input and was prone to becoming outdated over time.
In April 2025, the company took a significant leap by enabling ChatGPT to automatically reference past chat context beyond the manually saved memories. Powered by a background process dubbed “dreaming,” the model curates relevant snippets from conversation history, allowing it to learn from natural dialogue without requiring users to flag every detail.
OpenAI has unveiled a groundbreaking, compute‑efficient memory architecture that builds upon its “dreaming” technology, dramatically enhancing ChatGPT’s capacity to retain and apply context over extended conversations.
Internal benchmarks reveal the upgrade boosts factual recall from 67.9 % in 2025 to an impressive 82.8 % in 2026, while alignment with user preferences climbs from 55.3 % to 71.3 %. Accuracy over time also sees a substantial jump, rising from 52.2 % to 75.1 %.
What sets this iteration apart is a user‑friendly memory summary page, allowing users to review, edit, or correct ChatGPT’s stored information and specify which topics the model should surface and when.
The enhanced memory system is now live for ChatGPT Plus and Pro subscribers in the United States, with a phased rollout planned for additional countries and eventually for Free and Go tiers in the coming weeks.
News Source: Neowin
Comments
Be the first to comment.