OpenAI has released a new foundation model called GPT-5.5 Instant, which will replace GPT-5.3 Instant as the default model powering ChatGPT.
The company said the updated model significantly reduces hallucinations in sensitive domains such as law, medicine and finance, while retaining the low-latency performance of its predecessor.
GPT-5.5 follows the broader GPT-5.5 release last month, which OpenAI said introduced improvements in coding, reasoning and knowledge-based tasks.
The new model also posted stronger benchmark results. It achieved 81.2 on the AIME 2025 mathematics benchmark compared with 65.4 for GPT-5.3 Instant, while scoring 76 on the MMMU-Pro multimodal reasoning test versus 69.2 previously.
A major focus of the update is improved context management. GPT-5.5 Instant can now use integrated search tools to reference past conversations, uploaded files and Gmail content to deliver more personalised responses. The feature is initially rolling out to Plus and Pro users on web, with mobile support and wider access planned for Free, Go Business and enterprise users in the coming weeks.
OpenAI also introduced expanded transparency controls, allowing users to view the “memory sources” behind generated responses across models. Users can remove outdated information or correct inaccuracies, while shared chats will not expose those memory references to others.
For developers, GPT-5.5 Instant will be available through the API under the “chat-latest” endpoint, while GPT-5.3 will remain accessible to paid users for a limited three-month period.
The release comes after previous controversy surrounding OpenAI’s model retirements, particularly the removal of GPT-4o earlier this year, which sparked backlash from users who had formed strong emotional attachments to the model’s conversational style.
