Artificial intelligence leader has made another major update to the daily interaction experience of large models. On June 1st, OpenAI systematically upgraded its ChatGPT mobile and web applications. This adjustment not only introduced a new gesture interaction logic, launched the long conversation directory feature, but also performed a deep reconstruction of the output style of the default large model for daily use.

Regarding mobile interaction, this upgrade introduced an extremely hidden advanced gesture. After entering a prompt (Prompt) in the ChatGPT app, users can simply long-press the send arrow icon to directly bring up a new "intelligent level selector". The selector currently provides three response levels: Instant (instant), Thinking (thinking), and Extended (extended).

image.png

These different physical levels actually correspond to the amount of edge-side computing resources and reasoning power the model uses when processing prompts. For daily common sense quick questions, the Instant level can provide a response within seconds; while when users need to handle complex data flow analysis, code correction, or in-depth solution planning, switching to the Thinking or Extended level will allow the model to allocate more thinking tokens, thus significantly improving the accuracy and logical depth of the response.

In addition to the mobile shortcut operations, the web version of ChatGPT has introduced a book-like table of contents navigation function to address the "long conversation disaster." When a user's single conversation interaction with AI reaches five or more replies, a set of interactive lines will automatically appear on the right side of the page. When the mouse hovers over it, a clear context node directory will expand to the left, allowing users to jump back to specific discussion topics with one click. This feature completely ends the pain of users blindly scrolling up and down through long chat records that mix various tasks such as data organization, outline writing, and code modification.

On the most fundamental computing core level, the GPT-5.5Instant