OpenAI expands ChatGPT memory to free users, doubles storage capacity for paid tiers
OpenAI is rolling out an upgraded memory system for ChatGPT that synthesizes context more efficiently across conversations. The company reduced compute requirements by approximately 5x, enabling it to offer the memory feature to free users for the first time while doubling storage capacity for Plus and Pro subscribers.
OpenAI expands ChatGPT memory to free users, doubles storage capacity for paid tiers
OpenAI announced June 4, 2026 that it is expanding ChatGPT's memory feature to free users for the first time while releasing an upgraded memory system for paid subscribers in the United States.
Compute efficiency gains enable free tier expansion
The company reduced the compute required to serve its "dreaming-based" memory system by approximately 5x, according to OpenAI's announcement. This efficiency improvement makes it practical to offer the feature to free users at scale.
Free users will receive access to the memory feature "over the coming weeks," OpenAI says. The system automatically tracks important details and preferences across conversations without requiring users to repeat context.
New memory architecture for paid users
Plus and Pro subscribers in the US began receiving access to a "more capable and compute-efficient memory architecture" starting June 4. The upgraded system addresses what OpenAI describes as "staleness, correctness, and scalability challenges" when serving hundreds of millions of users over multi-year timeframes.
The new architecture includes a memory summary page where users can review what ChatGPT knows about them, add or update information, and provide instructions on which topics the system should reference and when.
OpenAI is also doubling memory storage capacity for Plus and Pro users, though specific storage limits were not disclosed.
Three focus areas for memory improvements
The upgraded system targets three specific capabilities:
- Carrying forward useful context across conversations
- Following preferences and constraints set by users
- Staying current with the passage of time
Users who prefer the previous saved memories experience can revert to the legacy system through settings.
What this means
The 5x compute reduction represents a significant engineering achievement that directly enables product expansion—free users gain a feature previously limited to paying customers, while paid tiers get expanded capacity. The move also signals OpenAI's focus on long-term context as a competitive differentiator, particularly as memory becomes table stakes for conversational AI systems that millions use daily. The phased US-only rollout for paid tiers suggests the company is stress-testing the new architecture before global deployment.
Related Articles
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.
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.
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.
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...