OpenAI Deploys GPT-5.5 Instant to Cut ChatGPT Hallucinations

OpenAI has made GPT-5.5 Instant the new default model powering ChatGPT. The company states that this update brings significant improvements in factuality across the board, addressing the persistent issue of hallucinations in AI responses.

According to OpenAI’s internal evaluations, GPT-5.5 Instant produced 52.5% fewer hallucinated claims than the Instant model for GPT-5.3. These tests focused on high-stakes prompts in areas like medicine, law, and finance. The new model also reduced inaccurate claims by 37.3% on challenging conversations that users had flagged for factual errors.

Beyond accuracy gains, GPT-5.5 Instant is described as more capable for everyday tasks. It handles image uploads more effectively and better determines when to consult the web for answers. Responses are tighter and more to the point, with fewer gratuitous emojis.

The model improves personalization by pulling context from previous chats and Gmail accounts more effectively. This allows ChatGPT to deliver responses tailored to individual users based on their history and connected data.

A new memory sources feature is being added to all ChatGPT models. It shows users exactly what context informed a personalized response. People can then delete or correct that information if desired.

GPT-5.5 Instant begins rolling out on Tuesday to all ChatGPT users. GPT-5.3 Instant will stay available as an option for three months before retirement. Enhanced personalization starts with Plus and Pro users on the web before expanding to mobile apps. OpenAI has plans to extend it soon to Free, Go, Business, and Enterprise users. The memory sources feature is rolling out now to consumer plans on the web and will reach mobile soon.