Internal testing indicates that GPT-5.5 Instant achieves 52.5% fewer hallucinated claims and 37.3% fewer inaccurate statements when addressing conversations flagged by users for factual errors. The model is designed to provide more direct answers while maintaining the warmth and personality that users appreciate.
OpenAI stated that the new model generates fewer unnecessary follow-up questions and offers less clutter in outputs, aided by a reduction in the use of emojis. Additionally, GPT-5.5 Instant has improved speed when searching uploaded files and connected accounts, such as Gmail, which should lead to a more seamless user experience.
Along with the model update, OpenAI is introducing a feature called memory sources. This feature allows users to see the context that ChatGPT uses for personalization and gives them control over their memory data. Users can delete or modify memory information they deem irrelevant, and this data will not be shared with others when a user chooses to share a chat.
OpenAI emphasized that users remain in control of their memory settings, with options to delete or change items as desired. The current version of memory sources may not capture all information used for responses, but improvements are planned for future iterations.
The rollout of GPT-5.5 Instant will begin today for all users, starting with Plus and Pro subscribers, followed by Free, Go, Business, and Enterprise users in the coming weeks. The memory sources feature will soon be available for all ChatGPT users on both web and mobile platforms.
