OpenAI claims ChatGPT’s new default model hallucinates way less
The Verge
OpenAI reports that its new GPT-5.5 Instant model significantly reduces hallucinations while enhancing task capability and personalization for ChatGPT users.Summary
OpenAI has introduced GPT-5.5 Instant, a new default model for ChatGPT designed to improve factual accuracy. Internal evaluations suggest a 52.5% reduction in hallucinations compared to the GPT-5.3 Instant model, particularly in complex fields like medicine, law, and finance. Beyond factuality, the model offers more concise responses, improved image analysis, and enhanced personalization by better utilizing context from past interactions and linked services like Gmail. Additionally, a new 'memory sources' feature allows users to view, manage, and delete information used by the chatbot for personalized responses. The model is rolling out to all users, with GPT-5.3 remaining available for a three-month transition period.
(Source：The Verge)