In this mode, Codex aims to use as few tokens as possible for a given task, while providing the bare minimum of working functionality. This is very useful for testing various ideas that can later be developed into more complex solutions.
This would be interesting. I think right now if I want to use “less” I tend to go with Codex spark. Have you used Codex spark? It’s decent for straight forward implementations or tasks that are well defined versus using something like 5.5 and extra high reasoning.
Thanks for the idea! You can ask Codex for a minimal working prototype with a tight scope: https://learn.chatgpt.com/docs/prompting. That isn't a dedicated minimum-token mode. We'll pass along that request; we don't have a timeline to share.