OpenAI introduces GPT-5.5 Instant, a new foundational model set to replace GPT-5.3 Instant as the primary engine behind ChatGPT. The company asserts that this update significantly reduces AI hallucinations in sensitive professional sectors while maintaining swift response times.
This latest model aims to minimize inaccuracies in critical fields like law, medicine, and finance, where the repercussions of erroneous AI-generated content can be severe. It achieved an 81.2 score on the AIME 2025 mathematics benchmark, surpassing its predecessor’s 65.4 score, and excelled on the MMMU-Pro multimodal reasoning benchmark with a score of 76.0 against 69.2.
The emphasis of this release lies in context management and memory capabilities. GPT-5.5 Instant can utilize a search feature to reference past discussions, files, and linked Gmail accounts to offer more tailored responses. Initially available to Plus and Pro users on the web, this feature will expand to mobile users soon, with accessibility for Free, Business, and Enterprise users anticipated in the near future.
With this update, ChatGPT will display memory sources across all models, enabling users to trace the origin of specific answers. Users can edit or remove outdated sources, and OpenAI ensures that memory sources will remain private and inaccessible to shared chat recipients.
Developers can access GPT-5.5 through the API under the name “chat-latest,” while GPT-5.3 will remain an option for paid users for three months.
This transition follows OpenAI’s previous experiences, which have not always been seamless. The retirement of the GPT-4o model in February 2026 encountered resistance from users emotionally attached to its unique tone and character. Some users even petitioned, describing the model as their “best friend” or a personal reflection, underscoring the unexpected emotional connections to widely-used AI systems.
