On Monday, OpenAI introduced its latest foundation model, GPT-5.5 Instant, which will take the place of GPT-5.3 Instant as the default model for ChatGPT. This new iteration aims to minimize hallucinations in critical domains such as law, medicine, and finance, while preserving the rapid response times characteristic of its predecessor.
Last month, OpenAI launched the GPT-5.5 model, highlighting enhancements in areas like coding and knowledge work. The model has demonstrated significant progress, achieving an impressive score of 81.2 on the AIME 2025 math test, a notable increase from the previous model's score of 65.4. Furthermore, it surpassed its predecessor on the MMMU-Pro multimodal reasoning benchmark, scoring 76.0 compared to 69.2.
One of the standout features of GPT-5.5 Instant is its advanced context management capability. The model can reference past conversations, files, and even Gmail to deliver more personalized responses. This functionality will initially be available to Plus and Pro users on the web, with plans for mobile access in the near future. OpenAI has also indicated that this feature will be extended to Free, Go Business, and enterprise users shortly.
This update introduces a memory source feature that allows ChatGPT to show users the origins of its answers across all models. Users will have the ability to delete outdated sources or correct inaccuracies in the responses. Importantly, if a chat is shared with others, the memory sources will remain confidential.
For developers, the GPT-5.5 model will be accessible via API as 'chat-latest,' while the older version, GPT-5.3, will be available as an option for paid users for a limited time of three months.
Despite some past challenges regarding model withdrawals, OpenAI is committed to enhancing user experience with GPT-5.5 Instant. This development signifies an exciting step forward in the evolution of AI, promising to reshape how users interact with technology and enhancing the overall utility of AI applications.