OpenAI rolled out GPT-5.5 Instant in May 2026, replacing GPT-5.3 Instant as the base model powering ChatGPT. The update is being made available to all users and brings improvements in accuracy, conciseness, and memory personalization.
The most notable change is a reported 52.5% reduction in hallucinations compared to GPT-5.3 Instant across domains including medicine, law, and finance. The model also produced 37.3% fewer inaccurate responses in conversations that users had previously flagged for factual errors. OpenAI says the update improves performance on everyday tasks such as answering STEM questions, analyzing image uploads, and determining when to use web search.
Responses are also shorter. In one example shared by OpenAI comparing communication advice from both models, GPT-5.5 Instant used 30.2% fewer words and 29.2% fewer lines than its predecessor. The company says the model reduces unnecessary follow-up questions and avoids excessive emoji use. “With this update, the model’s responses are tighter and more to-the-point without losing substance,” OpenAI wrote in its blog post.
Memory handling has also been updated. GPT-5.5 Instant is described as more effective at pulling context from past conversations, uploaded files, and connected Gmail accounts. The model is said to decide when additional personalization would improve a response and searches past conversations more quickly, potentially reducing the need for users to repeat themselves.
OpenAI is also introducing a “memory sources” feature across all its ChatGPT models. When the chatbot personalizes a response, it will display which past context or saved memory it referenced. Users can delete or correct that memory if it is outdated. Memory sources will not be visible to others if a chat is shared.
One outstanding question is whether a separate response quality issue — which OpenAI recently acknowledged had affected its GPT-5.5 model — also carries over into GPT-5.5 Instant. The company has not clarified this.
Source: mint – technology