I used api gpt 5.5 to review a python project that contain also venv folder. gpt 5.5 depleted all my tokens in 10 seconds scanning venv folder where he should know about it and skip for scanning. He is too much stupid that 5.4 or other AI’s
If I give a drive or windows he will scan completely the windows folder for bugs?
the API would not scan your folder, it only responds to discrete queries.
Are you talking about one of OpenAI’s applications or harnesses (e.g. Codex, Codex CLI) or a third party one like Cursor?
If so how are you authenticating? Via your API account?
And one other point - if you used a coding harness then I’m not a python expert (I’m a Rails/JS dev most of my days), but I would expect the harness to need to traverse and read the third party libraries in your venv to accrue information about their capabilities and APIs in order to code against them and deliver your spec.
You might be able to get ok quality from just reading your source, but I think that additional access will likely improve performance of the harness and model and it’s probably trained to do that. This might actually reduce the overall token usage as it will be better grounded in library capabilities.
5.5 is simultaneously the most expensive, but also the most capable coding model they have released.
So I think this is the cost of doing business now.