Feedback on GPT-5 in Cursor
After using GPT-5 in Cursor for the past few days, here are my conclusions regarding its integration with Cursor:
Overall, it performs well when using tools, except for:
- It does not use TODO lists, whereas O3 knew how to work with them.
- Occasionally (not often, but still) it writes that it has made changes when in fact it has done nothing. I am not sure what exactly causes this.
- It needs to be pushed to use the terminal and does not tend to go there on its own..
- There is no “no-thinking” model available. This is greatly missed, as on OpenAI’s website there is a model that responds quickly without thinking, but through Cursor there are only models that think extensively before responding.
These shortcomings are especially noticeable because they are completely absent in Anthropic.
That being said, it is an amazing model that I am very satisfied with, and I currently use it exclusively.
A huge thank you to the Cursor team for making this wonderful integration possible. Your work is truly appreciated, and the value it brings to my daily workflow is immense. I’m grateful for the effort, vision, and dedication that made this tool a reality.