Not true with github copilot. Cost is per prompt no matter how many tokens the prompt uses. Which can vary by 2 or 3 orders of magnitude...