OpenAI has released a default model for ChatGPT called GPT-5.5 Instant, replacing GPT-5.3 Instant. The company claims that the new AI model reduces hallucination and offers more accurate responses in day-to-day usage. This model is said to enhance daily interactions with the AI chatbot and claims to have a significant overall impact as it serves a large number of users every day.

The GPT 5.5 Instant is said to offer more accurate and refined responses across topics. It may offer a more human-like and natural interaction with improved personalisation based on the user’s context.

Must read: OpenAI launches ChatGPT for Intune app for secure work and study use

GPT‑5.5 Instant performance and improvements

According to OpenAI’s press note, the GPT‑5.5 Instant offers 52.5% fewer hallucinated claims than GPT‑5.3 Instant. Its inaccuracies have also been reduced by 37.3%, and the improvements have been made based on users’ feedback. 

OpenAI said, “GPT‑5.5 Instant is a generally smarter model that’s more capable across everyday tasks, including improvements in analysing photo and image uploads, answering STEM-related questions, and deciding when to use web search to provide a more useful answer.”

In terms of performance, the new GPT-5.5 Instant model also excels in benchmark testing. On the AIME 2025 math benchmark, the model scored 81.2 points, up from 65.4 points previously. On the MMMU-Pro multimodal reasoning benchmark, it scored 76 points, improving from 69.2 points earlier. Therefore, the model performs well when it comes to mathematical calculations, logical reasoning, and multimedia inputs.

Must read: OpenAI to roll out GPT-5.5-Cyber with restricted access: Sam Altman

Memory Source features for OpenAI’s AI models

The GPT 5.5 Instant also comes with the ability to draw context from past chats, files, and Gmail. This way, users will get personalised and contextual responses. In addition, OpenAI is rolling out a new feature called “memory sources” that gives users context of memories or past conversations that influenced that response. Users can also review that information and remove, edit, or correct anything that is outdated or no longer relevant.

Users will also have control over what they want to keep as Memory Sources, and can delete chats, edit saved memories, or use temporary chats that do not save or use memory at all.

Must read: OpenAI takes Codex to Amazon cloud, what’s next for Microsoft partnership?