I’m using gpt 5 with cline, and I’m loving it so far. This request is being made in the hopes of squeezing even more performance out of an already good model:
I’ve noticed that when creating new code gpt 5 likes to use short name variables, and I’d assume that’s intentionally done to keep the context window down… so far so good… but when it has to deal with my code which is quite often written to be self explanatory and verbose, I observe the token usage increasing quickly.
Would it be possible to create a minified representation of the files it’s working with cached locally on my machine, with each file representation being version controlled via git commit hash? I’d be perfectly fine with the extra time it takes to build up this compressed context if it better enables gpt 5 to work with my large codebases.