Yeah i noticed that too, i tried telling it that it had the tool but it still didn’t use it. IT also doesn’t use terminal commands or lots of other commands unless i tell it and then it does.
I’ve been testing it and it’s not bad, similar to gemini in attitude. Never had any model that doesn’t make mistakes, so i’m used to it. This one makes mistakes but it hasn’t gone crazy on me yet (like claude sometimes goes down the craziest rabbit holes doing stuff i didn’t really want, not that it’s totally bad stuff but just not necessary).
So far GPT has just stayed on track with exactly what i wanted so that’s at least saying something, even if it didn’t do it perfectly. However it’s request happy, it likes to stop and validate things as it goes, which honestly i like that but then it uses a request every time so that will add up. Since it’s free now i don’t care but if it keeps doing that (like limited amount of things it can do in one request) then it will be extremely expensive.
GPT-5 tools calls seem to generate tons and tons of errors of the forms “Tool call timed out after 10s (read file)”, “Tool call timed out after 10s (list files)”, “Tool call timed out after 12s (codebase search)”.
Doesn’t seem to be the issue with other models, such as claude 4 sonnet.
Steps to Reproduce
Ask GPT-5 any broad question or give a broad task about a large codebase
Operating System
MacOS
Current Cursor Version (Menu → About Cursor → Copy)