OpenAI has released GPT-5.3 Instant, with a roughly 27% reduction in high-risk scenario hallucination rate compared to the default 5.2 Instant.

According to 1M AI News, OpenAI has released GPT-5.3 Instant, replacing GPT-5.2 Instant as the default model for ChatGPT. This update focuses on improving everyday conversation experience, with main enhancements in three areas: reducing unnecessary rejections and preachy introductions, enhancing the quality of internet-sourced answer integrations, and reducing hallucination rates.

Based on OpenAI's internal evaluations, in high-stakes domains such as medicine, law, and finance, the hallucination rate when connected to the internet has decreased by 26.8% compared to the previous version, and by 19.7% relying solely on internal knowledge. In typical error-prone dialogues reported by users, the online hallucination rate decreased by 22.5% and the offline rate decreased by 9.6%. OpenAI stated that the tone of GPT-5.2 Instant was sometimes "cringe," making inappropriate assumptions about user intent or unnecessary statements, which version 5.3 targeted with specific adjustments.

This model is now available to all ChatGPT users, with the API endpoint provided as gpt-5.3-chat-latest. GPT-5.2 Instant will remain in the model list for paid users until retirement on June 3.