OpenAI has released GPT-5.5 Instant, a new foundation model that will replace GPT-5.3 Instant as the default model powering ChatGPT, with the company saying the update significantly reduces the rate of AI hallucinations in sensitive professional domains while maintaining low response times.
The new model is designed to cut down on inaccurate outputs in areas such as law, medicine, and finance — fields where the consequences of incorrect AI-generated information can be significant. It achieved a score of 81.2 on the AIME 2025 mathematics benchmark, compared with 65.4 for its predecessor, and outperformed the older model on the MMMU-Pro multimodal reasoning benchmark, scoring 76.0 against 69.2.
The release places particular emphasis on context management and memory. GPT-5.5 Instant can use a search tool to refer back to past conversations, files, and connected Gmail accounts to provide more personalised answers. This feature will be available initially to Plus and Pro users on the web, with a rollout to mobile planned shortly after, and access for Free, Business, and Enterprise users expected in the coming weeks.
With the update, ChatGPT will also show memory sources across all models, allowing users to understand where a given answer was drawn from. Users can delete outdated sources or correct errors, and the company confirmed that memory sources will not be visible to anyone a chat is shared with.
For developers, GPT-5.5 will be available through the API as "chat-latest," with GPT-5.3 remaining accessible as an option for paid users for a period of three months.
The transition follows a pattern that has not always been smooth for OpenAI. When the company retired its GPT-4o model in February 2026, it faced notable public resistance from users who had developed a strong attachment to the model's particular tone and personality. Some signed petitions describing it as their "best friend" or a form of personal mirror, highlighting the sometimes unexpected emotional dimensions of retiring widely used AI systems.
