OpenAI improves the ChatGPT that almost everyone uses

OpenAI is replacing the default model of ChatGPT, thereby changing the version of the chatbot that most users interact with daily. With now more than 900 million weekly active ChatGPT users, the update is more than just a technical detail.

According to internal tests, GPT-5.5 Instant is said to deliver around 52 percent fewer hallucinations in particularly critical requests from fields such as medicine, law, and finance compared to the previous model. In particularly difficult conversations marked by users as error-prone, the number of inaccurate statements has also decreased by about 37 percent, OpenAI claims in the announcement of the new model. Furthermore, the company describes GPT-5.5 Instant as being overall more capable in everyday use, for example, in analyzing photos and images, in answering scientific and technical questions, and in deciding when a web search is useful for a better answer.

GPT-5.5 Instant replaces the previous default model GPT-5.3 Instant, while GPT-5.5 Thinking, introduced just under two weeks ago, continues to be intended for more demanding use cases such as programming, research, and multi-stage analyses.

More insight into ChatGPT's memory

The way ChatGPT communicates is also said to have been revised, with a more natural tone and more concise answers. On average, OpenAI speaks of about 26 percent fewer words and 36 percent fewer lines. Additionally, the model asks fewer unnecessary follow-up questions and refrains from using superfluous emojis.

At the same time, GPT-5.5 Instant is said to incorporate existing context from previous chats, files, or Gmail more strongly into its responses, making them appear more personal. The model is intended to decide for itself when additional context is useful and to access relevant previous content faster than GPT-5.3 Instant.

With the new “Memory Sources” feature, ChatGPT users can, to a limited extent, see what context was accessed and influence it, for example, by deleting previous chats or adjusting saved memories. Users with data privacy concerns can also use temporary chats that do not save or update context. When sharing a chat, the context information used for personalization remains hidden.

GPT-5.5 Instant is currently being rolled out to all ChatGPT users. For paying users, the previous default model will remain optionally available for another three months before it is discontinued. The improved contextualization will initially reach Plus and Pro users on the web but is expected to be available on mobile devices soon and will be extended to other plans in the coming weeks. The “Memory Sources” feature will initially be rolled out to all ChatGPT users on the web and will follow on mobile devices shortly.

(olb)