OpenAI makes GPT-5.5 Instant the default model of ChatGPT

OpenAI makes GPT-5.5 Instant the default ChatGPT engine, cutting hallucinations by 52.5% and boosting scientific reasoning on GPQA from 78.5% to 85.6%.

GPT-5.5 Instant replaces GPT-5.3 Instant as ChatGPT's default engine for all users and simultaneously arrives in the API under the identifier chat-latest. The model is presented as more accurate and more concise, with responses better tailored to personal context when relevant.

Regarding factual accuracy, OpenAI claims 52.5% fewer hallucinated statements compared to the previous version on high-stakes prompts (medicine, law, finance), and 37.3% fewer errors in problematic conversations reported by users. Internal evaluations show notable progress in scientific reasoning (GPQA goes from 78.5% to 85.6%), competitive mathematics (AIME 2025 from 65.4% to 81.2%), expert multimodal reasoning (MMMU-Pro from 69.2% to 76%), and analysis of scientific diagrams (CharXiv-reasoning from 75% to 81.6%). The response format is also more concise: fewer words, fewer superfluous follow-up questions, fewer emojis, and less overloaded formatting.

OpenAI simultaneously introduces memory sources, accessible from all ChatGPT models: a panel that details the context leveraged to personalize a response (saved memories, past chats) with the option to correct or delete it. Enhanced personalization from past chats, files, and a connected Gmail account is first deployed on ChatGPT Plus and Pro on the web, before rolling out to mobile and extending to Free, Go, Business, and Enterprise offerings. GPT-5.3 Instant remains accessible to paid subscribers for three months via configuration settings before permanent removal.