{"id":105949,"date":"2026-06-05T14:02:38","date_gmt":"2026-06-05T10:02:38","guid":{"rendered":"https:\/\/techxmedia.com\/en\/?p=105949"},"modified":"2026-06-05T14:02:39","modified_gmt":"2026-06-05T10:02:39","slug":"chatgpt-memory-system-gets-major-upgrade","status":"publish","type":"post","link":"https:\/\/techxmedia.com\/en\/chatgpt-memory-system-gets-major-upgrade\/","title":{"rendered":"ChatGPT Memory System Gets Major Upgrade"},"content":{"rendered":"\n

OpenAI rolls out a new ChatGPT<\/a> memory system designed to improve scalability, accuracy, and long-term personalization for hundreds of millions of users.<\/p>\n\n\n\n

New Memory System Launched<\/strong><\/p>\n\n\n\n

OpenAI<\/a> has begun rolling out a more capable and scalable system for synthesizing memory in ChatGPT. The update is designed to address staleness, correctness, and scalability challenges that emerge when memory is used across hundreds of millions of users and multi-year time horizons.<\/p>\n\n\n\n

ChatGPT memory helps the system learn user preferences, projects, and constraints. As a result, future conversations can start with shared context instead of starting from scratch.<\/p>\n\n\n\n

Importantly, OpenAI said memory has become a core part of the ChatGPT experience. It helps the model understand users better over time. This enables more useful and relevant assistance.<\/p>\n\n\n\n

Rollout Availability<\/strong><\/p>\n\n\n\n

The update is currently available to Plus and Pro users in the United States. However, OpenAI confirmed that it will expand to additional countries in the coming weeks.<\/p>\n\n\n\n

Furthermore, the rollout will extend to Free and Go users over time.<\/p>\n\n\n\n

Evolution of ChatGPT Memory<\/strong><\/p>\n\n\n\n

Memory in ChatGPT first launched in April 2024. It was previously known as saved memories.<\/p>\n\n\n\n

At that stage, users had to explicitly ask ChatGPT to remember information. For example, a user might say, \u201cremember I am traveling to Singapore in July.\u201d However, this system relied heavily on direct instructions and often missed context.<\/p>\n\n\n\n

As a result, saved memories could feel limited and sometimes outdated.<\/p>\n\n\n\n

Introduction of \u201cDreaming\u201d<\/strong><\/p>\n\n\n\n

In April 2025, OpenAI introduced an update that allowed ChatGPT to reference chat context beyond saved memories. This system was called \u201cdreaming.\u201d<\/p>\n\n\n\n

Dreaming works by automatically curating memory from chat history in the background. Therefore, ChatGPT can learn from multiple conversations without explicit instructions.<\/p>\n\n\n\n

However, OpenAI noted that dreaming alone was not sufficient as a standalone memory system.<\/p>\n\n\n\n

New Memory Architecture<\/strong><\/p>\n\n\n\n

Now, OpenAI is launching a more advanced and compute-efficient memory architecture built on top of dreaming.<\/p>\n\n\n\n

In addition, the system includes a memory summary feature. This summary allows users to view what ChatGPT knows about them. It also enables editing, correction, or removal of stored details.<\/p>\n\n\n\n

Users can also manage instructions on what topics ChatGPT should remember or prioritize.<\/p>\n\n\n\n

How ChatGPT Memory Is Evaluated<\/strong><\/p>\n\n\n\n

OpenAI evaluates ChatGPT memory using three main objectives.<\/p>\n\n\n\n

First, it must carry forward useful context across conversations.<\/p>\n\n\n\n

Second, it must follow user preferences and constraints consistently.<\/p>\n\n\n\n

Third, it must stay current over time as real-world context changes.<\/p>\n\n\n\n

For example, temporary situations such as travel plans should update automatically when they are no longer relevant.<\/p>\n\n\n\n

Performance Improvements<\/strong><\/p>\n\n\n\n

OpenAI reports major improvements across versions of ChatGPT memory.<\/p>\n\n\n\n

For carrying forward context, performance increased from 41.5 percent in 2024 to 82.8 percent in 2026.<\/p>\n\n\n\n

For preference adherence, results improved from 31.4 percent to 71.3 percent.<\/p>\n\n\n\n

For staying current over time, performance rose significantly from 9.4 percent to 75.1 percent.<\/p>\n\n\n\n

These improvements reflect progress from saved memories in 2024 to dreaming V0 in 2025 and dreaming V3 in 2026.<\/p>\n\n\n\n

Real-World Use Cases<\/strong><\/p>\n\n\n\n

For example, ChatGPT can now remember a user\u2019s technical setup and provide more precise recommendations in future conversations.<\/p>\n\n\n\n

Without memory, responses remain generic and require repeated explanations.<\/p>\n\n\n\n

With memory, ChatGPT can tailor answers based on prior context such as tools, preferences, or constraints.<\/p>\n\n\n\n

Similarly, preferences like dietary choices or travel habits are now applied more consistently in responses.<\/p>\n\n\n\n

Scalability and Future Direction<\/strong><\/p>\n\n\n\n

OpenAI also highlighted efficiency gains in the new system. The compute required to serve dreaming has been reduced by approximately five times.<\/p>\n\n\n\n

This reduction allows memory to scale to Free users while maintaining quality standards.<\/p>\n\n\n\n

In addition, it supports increased memory capacity for Plus and Pro users.<\/p>\n\n\n\n

Conclusion<\/strong><\/p>\n\n\n\n

OpenAI says the new memory architecture provides a stronger foundation for personalization at scale. It aims to make ChatGPT more context-aware, consistent, and up to date across long-term usage.<\/p>\n\n\n\n

Ultimately, ChatGPT continues to evolve into a more personalized AI assistant through improved memory, scaling across all user tiers while maintaining accuracy and relevance.<\/p>\n","protected":false},"excerpt":{"rendered":"

OpenAI rolls out a new ChatGPT memory system designed to […]<\/p>\n","protected":false},"author":8,"featured_media":105951,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"inline_featured_image":false,"_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"footnotes":""},"categories":[9617],"tags":[9811,9812],"contributor":[9732],"class_list":["post-105949","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-latest-news","tag-chatgpt","tag-openai","contributor-news-desk"],"featured_image_src":"https:\/\/techxmedia.com\/en\/wp-content\/uploads\/2026\/06\/ChatGPT.jpg.jpeg","author_info":{"display_name":"Rabab","author_link":"https:\/\/techxmedia.com\/en\/author\/rabab\/"},"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/techxmedia.com\/en\/wp-json\/wp\/v2\/posts\/105949","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/techxmedia.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/techxmedia.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/techxmedia.com\/en\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/techxmedia.com\/en\/wp-json\/wp\/v2\/comments?post=105949"}],"version-history":[{"count":1,"href":"https:\/\/techxmedia.com\/en\/wp-json\/wp\/v2\/posts\/105949\/revisions"}],"predecessor-version":[{"id":105950,"href":"https:\/\/techxmedia.com\/en\/wp-json\/wp\/v2\/posts\/105949\/revisions\/105950"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/techxmedia.com\/en\/wp-json\/wp\/v2\/media\/105951"}],"wp:attachment":[{"href":"https:\/\/techxmedia.com\/en\/wp-json\/wp\/v2\/media?parent=105949"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/techxmedia.com\/en\/wp-json\/wp\/v2\/categories?post=105949"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/techxmedia.com\/en\/wp-json\/wp\/v2\/tags?post=105949"},{"taxonomy":"contributor","embeddable":true,"href":"https:\/\/techxmedia.com\/en\/wp-json\/wp\/v2\/contributor?post=105949"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}