OpenAI's new default ChatGPT model hallucinates less

OpenAI has updated GPT-5.5 Instant, the default model behind ChatGPT, and the changes are mostly about reliability rather than raw capability. OpenAI says the model produces 52.5 percent fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts in areas like medicine, law, and finance, which is the kind of prompt where a confident wrong answer does the most damage. It also writes more tightly, with about 30 percent fewer words on average, while keeping the conversational tone people expect from the default.

Two changes deserve more attention than the quality bump. OpenAI is adding memory sources across all ChatGPT models, a control that shows which past chats, files, or connected accounts were used to personalize a response, and lets you delete or correct that context. Separately, this is the first Instant model OpenAI classifies as High capability in both cybersecurity and biological and chemical preparedness, so it ships with extra safeguards rather than the lighter treatment earlier Instant models received. The announcement is on OpenAI's site.

Why it matters

Most people use the default model without choosing it, so a real cut in hallucinations on medical, legal, and financial questions reaches a lot of users at once. The memory controls also give you a concrete way to see and remove what ChatGPT is carrying about you between chats.