OpenAI has begun rolling out an upgraded memory feature to users of the free version of ChatGPT, a tool that was previously exclusive to those on the Pro plan. The enhancement allows the AI chatbot to provide more tailored responses by recalling user preferences and recent conversations.
Announced in April and gradually being extended to more users, the updated memory system enables ChatGPT to offer improved support in activities such as writing, learning, and general advice by referencing recent interactions. OpenAI confirmed the development on X, stating, “We’re starting to roll out a lightweight version of memory improvements to Free users. In addition to existing saved memories, ChatGPT now references your recent conversations to provide more personalised responses.”
Users can test how much the chatbot remembers by simply prompting it to “Describe me based on our chats.” While the model may not recall interactions from several months ago, it does retain information from more recent sessions.
For those who prefer not to use the memory function, OpenAI offers options to disable it. The setting can be found under Settings > Personalisation > Memory, where users can switch off memory entirely or review and delete previously stored information.
In parallel, OpenAI has also expanded access to its Codex coding assistant. Initially available only to Enterprise, Team, or Pro subscribers, the tool is now accessible to ChatGPT Plus users. Codex is designed to support developers with writing, debugging, and testing code.
A significant new feature in this update is Codex’s ability to access the internet during programming tasks. With this function enabled, the assistant can install packages, interact with staging environments, and fetch resources needed for running specific tests. However, internet access is disabled by default and must be manually activated. Users can also control which domains Codex may reach and what HTTP methods it is allowed to use.
OpenAI stated it is monitoring the system for any potential misuse, including so-called “prompt injection attacks,” to ensure the coding assistant operates securely and as intended.
