product updateOpenAI

OpenAI upgrades ChatGPT memory architecture with automatic 'dreaming' synthesis, now available to free users

TL;DR

OpenAI is rolling out a new memory architecture for ChatGPT that automatically synthesizes information across conversations without explicit user prompts. The company announced free tier users will access memory features for the first time, while Plus and Pro users receive expanded memory capacity.

2 min read
1

New Memory Architecture Replaces Basic Saved Memories

OpenAI is rolling out a new memory architecture for ChatGPT that automatically synthesizes information across conversations without explicit user prompts. The company announced free tier users will access memory features for the first time, while Plus and Pro users receive expanded memory capacity.

The update builds on OpenAI's "dreaming" feature, a background process that has been supplementing basic saved memories since 2024. According to OpenAI, the original saved memories feature launched in April 2024 required strong user cues and suffered from decreasing relevance over time.

How the New System Works

The updated architecture generates a "memory summary" that users can read, edit, and manage at any time. Users can add information about themselves and specify when ChatGPT should reference stored context.

According to OpenAI, the system now better carries forward context across conversations. In one example, if a user previously discussed their camera equipment, ChatGPT will generate product recommendations compatible with that specific setup in future conversations. The system also automatically revises memories over time — preventing the chatbot from referencing past trips as upcoming events.

The new architecture works alongside the memory sources feature released with GPT-5.5 Instant, which allows users to see and edit the specific information ChatGPT used to personalize responses.

Free Tier Gets Memory for First Time

OpenAI attributes free tier access to "behind-the-scenes efficiency improvements" that make the dreaming process computationally cheaper to run at scale. The company states this marks the first time free account holders will have access to memory features through the dreaming process.

Plus and Pro subscribers receive greater memory capacity compared to free users, though OpenAI did not specify exact storage limits for either tier.

The rollout begins today for Plus and Pro users in the United States, with international availability planned for "the coming weeks." Free tier access timeline was not disclosed.

What This Means

This update signals OpenAI's focus on stateful AI interactions as a competitive differentiator. By extending memory to free users, the company is normalizing persistent context as a baseline chatbot feature rather than a premium capability. The efficiency gains that enabled free tier access suggest OpenAI has found ways to compress or optimize memory operations — potentially through architectural changes or more efficient retrieval mechanisms. The automatic revision of memories to maintain temporal accuracy addresses a core weakness of earlier implementations where stale context degraded user experience.

Related Articles

product update

Pentagon Adds OpenAI's ChatGPT Mil and xAI's Grok for Government to GenAI.mil

The Pentagon has added OpenAI's ChatGPT Mil and xAI's Grok for Government to its GenAI.mil platform, which previously offered only Google Gemini. Anthropic's Claude remains excluded after a supply-chain risk dispute with the Trump administration.

research

OpenAI's Reported 'Opaque Recurrence' Technique in Upcoming Astra Model Alarms AI Safety Researchers

The Information reports OpenAI's upcoming Astra model uses 'recurrent depth,' or 'opaque recurrence,' a technique that processes queries in loops rather than linear steps. AI safety researchers, including Redwood Research's Buck Shlegeris and Ryan Greenblatt, warn the approach could erode chain-of-thought monitorability if scaled further.

analysis

Safety Researchers Warn OpenAI's Unreleased Astra Model May Hide Its Reasoning From Monitors

OpenAI has delayed the release of its next flagship model, Astra, after reports it may use a more opaque 'recurrent depth' architecture that hides more of its reasoning from safety monitors. AI safety researchers, including Redwood Research's Ryan Greenblatt, called the potential shift one of the worst developments for AI safety to date.

model release

OpenAI Rates Upcoming Astra Model 'Critical' Risk for Cyber Capabilities — Its Highest Tier Ever

OpenAI says its unreleased Astra model is the first to trigger a 'critical' cybersecurity rating under its Preparedness Framework, capable of finding and chaining unknown vulnerabilities without human guidance. The company calls it simultaneously its most dangerous and safest model, while a new architecture detail raises questions about how well its reasoning can still be monitored.

Comments

Loading...