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

OpenAI has released GPT-5.5 Instant on Monday, officially replacing GPT-5.3 Instant as the default model for ChatGPT. The company stated that this update specifically aims to reduce hallucinations within sensitive domains, including law, medicine, and finance, without compromising the low latency that characterises the Instant version. This strategic shift follows the broader release of the GPT-5.5 model last month, which focused on improvements in coding and knowledge work.

The new iteration demonstrates significant gains in performance metrics compared to its predecessor. In the AIME 2025 mathematics test, GPT-5.5 Instant achieved a score of 81.2, a marked improvement over the 65.4 recorded by GPT-5.3. Furthermore, the model outperformed the previous version on the MMMU-Pro multimodal reasoning benchmark, securing a score of 76.0 against a 69.2 rating. These benchmarks suggest a substantial leap in the model's reasoning capabilities across complex tasks.

A key component of the update is enhanced context management, which allows the model to utilise search tools to reference past conversations, uploaded files, and Gmail data to provide more personalised answers. This feature is initially available to Plus and Pro users on the web, with a planned rollout to mobile devices in the near future. OpenAI has indicated that access for Free, Go Business, and enterprise users is expected in the coming weeks.

To increase transparency, the update introduces memory sources across all models, displaying the origins of generated answers to the user. This feature enables users to delete or correct sources if an answer is found to be incorrect. Crucially, OpenAI confirmed that these memory sources are hidden when a chat is shared with others, balancing transparency with user privacy.

For developers and enterprise integrations, the GPT-5.5 model is accessible via API under the identifier 'chat-latest'. Meanwhile, the GPT-5.3 version will remain an option for paid users for a limited period of three months. This transition comes after OpenAI previously faced significant backlash in February 2026 when it withdrew the GPT-4o model, a move that prompted petitions from users who described the model as a "best friend" or a "mirror".

While the company focuses on software updates, industry reports suggest OpenAI is also fast-tracking the development of a smartphone for potential mass production in early 2027. However, this remains unconfirmed by the company, and investors will continue to monitor official statements regarding any hardware initiatives.