OpenAI has begun deploying its updated ChatGPT default model, GPT-5.5 Instant, following the recent release of GPT-5.5. The new model is designed to produce 52.5% fewer hallucinated claims and 37.3% fewer inaccurate claims in conversations identified by users as containing factual errors.
According to OpenAI, GPT-5.5 Instant generates more direct responses while maintaining a warm and personable interaction style. The system is engineered to ask fewer unnecessary follow-up questions and produce less cluttered outputs, in part by using emojis more sparingly. Additionally, the model has enhanced capabilities for searching uploaded files, linked Gmail accounts, and previous ChatGPT conversations to create personalized responses, potentially reducing the frequency with which users need to repeat themselves.
OpenAI is also introducing a feature named memory sources, which provides users with visibility over the context used for personalizing responses. This feature allows users to delete or update information in memory sources that may be obsolete. “Memory sources aren’t shown to others if you choose to share a chat,” OpenAI noted. Users maintain control over their memory, with the ability to delete chats or modify items in saved memories.
The current version of memory sources does not reveal all information that might have been used in generating answers, although OpenAI is actively working to improve its comprehensiveness. The rollout of GPT-5.5 Instant will initially target Plus and Pro subscribers, with Free, Go, Business, and Enterprise users expected to gain access in the near future. OpenAI indicated that all ChatGPT users will receive the memory sources feature on both web and mobile platforms “soon.” The deployment of GPT-5.5 Instant commenced today, marking a significant update for the platform.
