OpenAI simply unveiled a big improve to ChatGPT’s reminiscence capabilities, enabling the AI to reference a person’s whole dialog historical past to ship extra customized responses. The characteristic, rolling out first to paying prospects of OpenAI’s Professional tier, marks a step towards extra customized AI techniques.
“Now we have vastly improved reminiscence in ChatGPT—it may possibly now reference all of your previous conversations!” OpenAI CEO Sam Altman introduced on X. “This can be a surprisingly nice characteristic imo, and it factors at one thing we’re enthusiastic about: AI techniques that get to know you over your life, and turn into extraordinarily helpful and customized.”
we’ve vastly improved reminiscence in chatgpt–it can now reference all of your previous conversations!
it is a surprisingly nice characteristic imo, and it factors at one thing we’re enthusiastic about: ai techniques that get to know you over your life, and turn into extraordinarily helpful and customized.
— Sam Altman (@sama) April 10, 2025
The enhancement expands past ChatGPT’s earlier reminiscence operate, which allowed customers to manually save particular data—and is already fairly good and correct. Now, the system can mechanically pull context from any earlier dialog to tailor its responses primarily based on a person’s demonstrated preferences, writing type, and recurring questions.
OpenAI claims customers keep management over their knowledge, with choices to choose out of the reminiscence characteristic completely or use momentary chats that will not contribute to or draw from the system’s reminiscence banks.
“If you wish to change what ChatGPT is aware of about you, merely ask in chat,” OpenAI famous in its announcement.
The characteristic arrives at a time when AI firms are racing to make their instruments extra useful by turning into extra private, and in a position to course of much more data. Think about with the ability to export your cloud of AI interactions to, say, a brand new humanoid robotic assistant who will “know” you as quickly because it powers on. (Or possibly don’t.)
The replace goals to create a extra steady relationship between the person and AI, and is one step in the identical route OpenAI took when releasing GPT-4.5—a mannequin that was constructed to be extra highly effective, but additionally to have a extra pure dialog type.
“New conversations naturally construct upon what it already is aware of about you, making interactions really feel smoother and uniquely tailor-made to you,” OpenAI defined in its announcement thread.
Altman hinted on the announcement earlier at this time. “A number of instances a yr I get up early and might’t fall again asleep as a result of we’re launching a brand new characteristic I have been so enthusiastic about for therefore lengthy. In the present day is a kind of days!”
His cryptic message sparked hypothesis amongst followers about potential new fashions or expanded capabilities. Some fanatics steered OpenAI may be launching new fashions like “o4-mini,” with some media shops reporting that it may even be a brand new GPT mannequin—which Altman denied. Nonetheless, when a developer hinted at the potential for OpenAI increasing its token context window to 1 million tokens—which is called Challenge Quasar—Altman replied saying that “quasars are very brilliant issues.”
It’s doable that a rise within the token context dealing with capabilities, alongside some highly effective RAG (retrieval augmented era) instruments, make it doable for the mannequin to course of all the data customers have saved up to now.
Based on an electronic mail shared by OpenAI with Decrypt, the mannequin can now reference chat historical past—although it doesn’t deal with them in the identical manner it handles reminiscences. “Reference chat historical past permits ChatGPT to attract context from previous conversations to adapt to your tone, objectives, pursuits, or different recurring matters. This evolves over time and isn’t saved or proven in settings the way in which saved reminiscences are,” the e-mail famous.
The reminiscence improve may even be obtainable to Plus subscribers someday later sooner or later. As normal, folks in a number of European areas—together with the European Financial Space, UK, Switzerland, Norway, Iceland, and Liechtenstein—will not get entry, seemingly as a result of area’s strict privateness laws.
Crew, Enterprise, and Training customers will acquire entry to the improved reminiscence options inside “a number of weeks,” in line with OpenAI.
Even earlier than this growth, ChatGPT’s current reminiscence capabilities have been already substantial. We requested ChatGPT to generate person profiles primarily based on previous interactions, and the system may establish jobs, character traits, relationships, and typically even delicate data shared in earlier conversations. That mentioned, regardless of the data coming from earlier chats, it’s nonetheless not in a position to reference previous conversations—which is the brand new replace being launched at this time.
OpenAI hasn’t specified precisely how lengthy ChatGPT will retain dialog historical past or how extensively it is going to mine that knowledge to construct person profiles. For these involved about privateness, the momentary chat choice gives a manner for customers to work together with the AI with out contributing to its rising understanding of their preferences and behaviors.
Usually Clever E-newsletter
A weekly AI journey narrated by Gen, a generative AI mannequin.