GPT-5.2 Codex is now available in Cursor!
We believe it’s the frontier model for long-horizon tasks.
3 Likes
is free ? short time ? or costed ?
1 Like
Oh, cheaper than GPT-5.2!
Really worth a try.
costed bro, but it’s cheap as composer-1 model
3 Likes
Unfortunately, the integration in Cursor is awful again: broken output (when every character or word is printed on a new line), very poor tool handling, and weak support for results produced by other models.
I don’t understand what you’re doing, but most likely you need to make a valid script for reading .env or its equivalent, through which Agent can execute your order.
1 Like
No, it is the model. Opus 4.5 for example have no problem what so ever taking my secret key and using it to run the dotnet command.
I think gpt 5.2 has explicit safety to not read or use the api keys, I faced it too. In short term security stuff might feel annoying, but in long term is often more valuable.
Blah, I’m not really interested on products which decide these things for me.
what are the rates for any of the dozen 5.2 Codex variants?
not sure if it is repeatable or not, but 5.2 codex fast clears context when confirming a plan then doesn’t have a to-do list
in my experience so far:
- codex 5.2 fast version is slower than the regular (?)
- price per task is same as opus
- quality comparable to opus
- so far codex does not do the weird bugs as it used to (using python for everything, disobeying instructions etc)
5.2 codex seems to do pretty well in Cursor, congrats!
1 Like