Suggestion: Reserving Usage

I’m all for the recent efficiency updates to Codex for Luna (no matter how much OpenAI’s actual costs dropped), but it's frustrating that heavy work with Sol can completely lock you out of lighter models. Once a big project or refactor drains your quota, you can't even use Luna for quick, simple tasks. It’d be great to see a fix for this, like a small, fallback 3–4% Luna-only allowance that kicks in after the main limit is hit, or a customisable reserve threshold that locks heavy usage (and cuts off chat continuation to prevent abuse) so users can save capacity for lightweight tasks.