GPT-5.5 Instant Becomes ChatGPT's Default Model

OpenAI has introduced GPT-5.5 Instant as the new default model for ChatGPT, replacing GPT-5.3 Instant for most everyday conversations. The update is positioned around a familiar but important promise: keep the speed of the Instant model line while making answers more reliable.

The company says GPT-5.5 Instant is designed to reduce hallucinations in sensitive areas such as law, medicine, and finance, where small factual errors can have outsized consequences. It also brings stronger benchmark performance, including a large jump on the AIME 2025 math test and gains on multimodal reasoning evaluations.

Better Context and More Transparent Memory

One of the bigger changes is how the model handles context. GPT-5.5 Instant can draw on past conversations, uploaded files, and connected services such as Gmail when search and memory features are enabled, giving ChatGPT more personalized context for follow-up questions and ongoing work.

OpenAI is also expanding memory-source visibility across ChatGPT. Users will be able to see which memories or prior context informed an answer, remove outdated sources, or correct information when the model relies on something inaccurate. Shared chats will not expose those private memory sources to other people.

What It Means for Users and Developers

For ChatGPT users, the practical effect should be a faster default assistant with better reasoning, more useful personalization, and clearer controls over remembered context. Plus and Pro users on the web are expected to see the new contextual features first, with mobile and broader account tiers following later.

For developers, GPT-5.5 will be available through the API under the chat-latest route. GPT-5.3 will remain available as an option for paid users for a limited transition period, giving teams a short window to test behavior changes before fully moving over.

The rollout also highlights the ongoing tension around model retirements. As AI assistants become more personalized, users increasingly notice differences in model behavior, tone, and reliability. OpenAI's challenge is to ship stronger defaults without making people feel that a familiar assistant has been abruptly swapped out.