OpenAI Releases GPT-5.5 Instant for ChatGPT With Improved Accuracy and Memory Controls

OpenAI has rolled out GPT-5.5 Instant, replacing ChatGPT's default model with smarter responses, improved accuracy, and better personalization. The update begins rolling out to all users today.



Accuracy in high-stakes topics is a primary focus of this release. OpenAI says GPT-5.5 Instant produced 52.5 percent fewer hallucinated claims than its predecessor on high-stakes prompts covering medicine, law, and finance. It also reduced inaccurate claims by 37.3 percent on conversations that users previously flagged for factual errors. The model is generally more capable across everyday tasks, bringing noticeable improvements to visual reasoning, math, and STEM-related questions. It is also designed to be smarter about deciding when to pull in live web search results.


The chatbot's tone and formatting have been refined to reduce clutter. Responses are now tighter and more direct, cutting back on overly verbose responses and heavy formatting that often made previous versions feel bloated. OpenAI says the model now asks fewer unnecessary follow-up questions and avoids clutter like gratuitous emojis, while still maintaining a natural conversational style.

OpenAI has also upgraded how ChatGPT handles context. The chatbot is now much faster at searching past conversations to find relevant details, meaning users will not have to repeat themselves as often. For eligible users, it can also pull context from uploaded files, past chats, and connected Gmail accounts to tailor suggestions. To make this process more transparent, OpenAI is introducing memory sources across all ChatGPT consumer plans on the web.

When personalization influences a response, users can now see what saved memories or past chats were referenced. If the AI pulls up outdated or irrelevant information, users can delete or correct those specific memory sources directly from the chat interface. For privacy, these memory sources are hidden if you choose to share a conversation link with someone else. Users can also utilize temporary chats that bypass the memory system entirely.

GPT-5.5 Instant replaces GPT-5.3 Instant, which OpenAI released earlier this year. The model is also now available in the API as chat-latest. Paid users will retain access to GPT-5.3 Instant through model configuration settings for three months before the model is retired. The new memory sources feature is rolling out on the web first and will reach mobile soon. Enhanced personalization capabilities are currently live for Plus and Pro users on the web, with plans to expand to the ChatGPT Go, Free, Business, and Enterprise tiers in the coming weeks. This consumer rollout follows the recent launch of the more advanced GPT-5.5 model geared toward complex coding and agentic workflows.