OpenAI has launched a new foundation model called GPT-5.5 Instant, replacing GPT-5.3 Instant as the default model powering ChatGPT.
The company says the updated model delivers faster and more reliable responses while reducing hallucinations in sensitive topics such as law, medicine, and finance.
The release comes just weeks after OpenAI introduced the full GPT-5.5 model, which the company said brought improvements in coding, reasoning, and knowledge-based tasks. According to OpenAI, GPT-5.5 Instant also performs significantly better on benchmarks compared to its predecessor.
In the AIME 2025 math benchmark, the new model scored 81.2, up from 65.4 achieved by GPT-5.3 Instant. It also improved its results on the MMMU-Pro multimodal reasoning benchmark, scoring 76 compared to the older model’s 69.2.
A major focus of the update is context and personalization. GPT-5.5 Instant can now use its search capabilities to reference past conversations, uploaded files, and Gmail content to generate more personalized responses. The feature is currently available for Plus and Pro users on the web, with mobile support planned soon.
OpenAI said it also plans to expand access to Free, Go Business, and enterprise users over the coming weeks. Alongside the rollout, ChatGPT will begin showing memory sources across all models so users can better understand where information in responses came from. Users will also be able to remove outdated sources or correct incorrect memories. OpenAI noted that shared chats will not expose memory sources to other people.
For developers, GPT-5.5 Instant will be available through the API under the name “chat-latest.” OpenAI also confirmed that GPT-5.3 will remain available for paid users for another three months before being retired.
If this article helped you, please consider supporting our work. Every small contribution keeps Abijita.com independent and running.
The company has previously faced criticism over retiring older models. When OpenAI removed GPT-4o earlier this year, many users objected because they had become emotionally attached to the model’s conversational style. Some users described GPT-4o as their “best friend” or “a mirror,” and petitions were created asking OpenAI not to discontinue it. Despite the backlash, GPT-4o was officially retired in February 2026.
