On Tuesday, OpenAI launched the new foundation model GPT-5.5 Instant, replacing GPT-5.3 Instant as the default in ChatGPT. The updated model is said to reduce inaccuracies in sensitive fields like law and medicine while retaining quick responses.
The latest release boasts improvements in coding and knowledge work, scoring 81.2 on AIME 2025 math tests compared to its predecessor's 65.4. It also outperformed the older model by 76 vs 69.2 on the MMMU-Pro benchmark.
A notable addition is enhanced context management: GPT-5.5 Instant can reference past conversations, files and Gmail for personalized answers, with this feature set to roll out to Plus and Pro users soon. Free users may get access in weeks.
Users will be able to see memory sources across all models, allowing corrections or deletions of outdated information. However, shared chats won't reveal these sources to others.
For developers, the new model is available as ‘chat-latest’ with GPT-5.3 still an option for paying users for three months.
