When you prompt the AI and while generating it gives you an error that you need to click refresh, it will take another fast request from you.
So from those 500 fast requests you could have used only 100-200 because the rest fast used by you clicking “try again” or “retry”.
From my own experience the AI crashing so often in the chat that i would sometimes use even 10 fast requests in 1 prompt, because the AI kept crashing.
Please fix it fast, give us more requests or at least specify this because right now it is misleading.