OpenAI has introduced GPT-5.5 Instant as the new default model for ChatGPT, replacing GPT-5.3 Instant in a move aimed at improving reliability without sacrificing speed. The company says the updated model significantly reduces hallucinations — particularly in sensitive domains such as law, medicine and finance — while maintaining the low-latency performance that made its predecessor suitable for everyday use.
In other words, the goal isn't just smarter answers, but fewer confidently wrong ones. The release follows last month's broader rollout of the GPT-5.5 family, where OpenAI claimed gains in coding and knowledge-based tasks. GPT-5.5 Instant appears to inherit some of those improvements while staying optimised for responsiveness — a balancing act that has become central to mainstream AI products.
On benchmarks, the gains are tangible. The model scored 81.2 on the AIME 2025 mathematics test, up from 65.4 for GPT-5.3 Instant. It also improved its performance on the MMMU-Pro multimodal reasoning benchmark, suggesting stronger capabilities across text and visual tasks.
Beyond raw performance, OpenAI is placing a clear emphasis on context and personalisation. GPT-5.5 Instant can draw on past conversations, uploaded files and even Gmail data through its search tool to generate more tailored responses. This feature is rolling out first to Plus and Pro users on the web, with mobile support and broader availability expected in the coming weeks.
AI Memory Gets More Transparent
Transparency is another focus area. ChatGPT will now display "memory sources" across models, allowing users to see where information in a response is coming from. Users can delete or correct these sources, a nod to growing concerns around AI traceability and control. Notably, shared chats will not expose these memory references to others, addressing a potential privacy gap.
For developers, GPT-5.5 will be accessible via API under the "chat-latest" label, while GPT-5.3 will remain available as a paid option for a limited three-month window — a transition period that hints at OpenAI's increasingly rapid upgrade cycles. The company faced backlash earlier this year when it retired GPT-4o, with some users expressing attachment to the model's tone and behaviour.
