OpenAI has launched an update to its ChatGPT memory feature to address historical accuracy issues. The update, named Dreaming V3, automates how the system synthesizes conversation history. Meanwhile, this background process aims to keep stored information current without requiring manual user commands.
The Evolution of ChatGPT memory
According to internal data published by OpenAI, the previous version of the memory system had a factual recall accuracy of 41.5% in 2024. Consequently, the system was incorrect in more than half of memory-dependent situations. The new update increases this recall rate significantly while reducing operational demands. Notably, this feature is now rolling out to Plus and Pro users in the United States, with free users receiving access in the coming weeks.
Technical Performance Improvements
Specifically, internal tests show that factual recall has risen to 82.8% in 2026. In addition, preference adherence, which tracks how well the system follows user style and habits, improved from 55.3% to 71.3%. The system also self-corrects over time, such as removing travel recommendations once a trip ends. Furthermore, compute costs dropped fivefold, enabling free users to access the service for the first time.
Integration with Financial Services
This update follows a May 2026 launch of a personal finance feature for Pro users. This tool allows users to connect their bank accounts to analyze spending habits. Combining ChatGPT memory with financial data creates a more personalized utility for managing daily tasks. Meanwhile, memory storage capacity for Plus and Pro users has doubled to accommodate this data. Users can manage these settings by navigating to their profile, selecting personalization, and viewing the memory summary.
Future Outlook for AI Assistants
The admission of the previous 41.5% accuracy rate highlights a broader challenge for the smart-devices and software industry. Most current digital assistants face similar recall issues without notifying users. Therefore, the development of ChatGPT memory represents a shift toward more transparent performance metrics in the consumer technology sector. As a result, users may expect more reliable interactions with generative models in the future.
