OpenAI shifts ChatGPT default to GPT-5.5 Instant, prioritising accuracy in high-stakes sectors

OpenAI has officially designated GPT-5.5 Instant as the new default model for ChatGPT, superseding the previous GPT-5.3 version. The update is scheduled to roll out to all users on Tuesday, marking a significant shift in the platform's core capabilities. While the transition follows a pattern of rapid iteration, OpenAI has ensured that GPT-5.3 Instant will remain available as an option for three months before being retired, allowing users time to adjust to the changes.

The primary driver behind this update is a marked improvement in factuality, addressing the ongoing issue of hallucinations that have plagued AI models. OpenAI states that internal evaluations show GPT-5.5 Instant produces 52.5% fewer hallucinated claims on high-stakes prompts covering critical areas such as medicine, law, and finance. Furthermore, the model reduces inaccurate claims by 37.3% on challenging conversations that have previously been flagged for factual errors. These metrics, while based on the company's internal assessments, highlight a concerted effort to enhance reliability in sectors where precision is paramount.

Beyond factual accuracy, the new model aims to deliver tighter, more direct responses by utilising fewer gratuitous emojis. The update also claims enhanced capability in everyday tasks, including better analysis of image uploads and improved judgment on when to consult the web for answers. This refinement in output style is designed to make interactions more efficient and less prone to the decorative fluff that can sometimes obscure information in previous iterations.

A key feature of the GPT-5.5 Instant rollout is the expansion of enhanced personalisation, which allows the model to pull context from previous chats and Gmail accounts. This functionality, which mirrors similar investments by competitors like Google for its Gemini model, is initially rolling out to Plus and Pro subscribers on the web. OpenAI indicates that these personalisation features are coming soon to mobile apps and plans to extend them to Free, Go, Business, and Enterprise users in the near future, though specific dates for these broader rollouts have not been confirmed.

To address concerns regarding transparency and user control, the update introduces a new 'memory sources' feature. This tool enables users to view the specific context used to inform personalised responses and provides the ability to delete or correct that information. The feature is currently rolling out to ChatGPT consumer plans on the web and is expected to arrive on mobile devices soon. This move represents a significant step towards giving users greater oversight over the data driving the AI's suggestions.

As the technology sector continues to race towards more reliable artificial intelligence, OpenAI's focus on reducing errors in high-stakes domains signals a maturing approach to the technology. While the specific definitions of 'high-stakes prompts' and 'challenging conversations' used to derive the reduction metrics are not explicitly detailed in the source text, the emphasis on accuracy suggests a strategic pivot towards trust and utility. The full update will be accessible to all users by Tuesday, setting a new baseline for the ChatGPT experience.