Krux
May 7, 2026
ChatGPT Cuts Hallucinations by Half on Critical Tasks
Published: May 7, 2026 at 12:13 AM
Updated: May 7, 2026 at 12:13 AM
100-word summary
OpenAI's GPT-5.5 Instant is now the default ChatGPT model, and the upgrade is significant: hallucinations drop 52.5% on high-stakes prompts compared to the previous version. The model also cuts inaccurate claims by 37% in tough conversations, a meaningful shift for anyone using ChatGPT in medicine, law, or finance where wrong answers have consequences. New personalization features pull context from your chat history, uploaded files, and even Gmail to tailor responses. Plus users get transparency controls to see exactly what shaped each answer. The model can now automatically switch to deeper reasoning mode when needed, or you can force it to think longer on complex problems.
What happened
OpenAI's GPT-5.5 Instant is now the default ChatGPT model, and the upgrade is significant: hallucinations drop 52.5% on high-stakes prompts compared to the previous version. The model also cuts inaccurate claims by 37% in tough conversations, a meaningful shift for anyone using ChatGPT in medicine, law, or finance where wrong answers have consequences. New personalization features pull context from your chat history, uploaded files, and even Gmail to tailor responses. Plus users get transparency controls to see exactly what shaped each answer.
Why it matters
The model can now automatically switch to deeper reasoning mode when needed, or you can force it to think longer on complex problems.