Auto model does not seem to be so good. I have used it for the last couple of days after hitting my usage limit on pro and can say it’s not very smart and in some cases it won’t understand user intent even when the prompt is clearly written. Cursor says auto model automatically routes requests to different (including premium) models based on task complexitiy. I have used for both small and large tasks and its quality has been the same. It’s always fast for even more complex tasks, making me believe that it doesn’t use thinking models such as gpt 5 or gpt codex. What do you guys think?