Jargon:Token-rationing
1 min read
Say this at the party
“Making your AI allowance last until it resets, by saving the big jobs for when you've got the most left.”
In plain English
Paid AI plans aren't truly unlimited. Claude, for example, gives you a session allowance that resets every five hours, with a weekly limit on top. Hit either and you wait.
Token-rationing is everything people do to avoid that wall: using a smaller model for easy questions, keeping chats short, batching questions into one message and timing heavy work for just after a reset.
It's the opposite of token-maxxing. One treats tokens as a scoreboard, the other as a budget.
Where it came from
As coding agents got hungrier through 2025 and 2026, AI companies put rolling session windows and weekly caps on their monthly plans. Heavy users started planning their days around reset times, and talk of rationing followed. It's a word from users, not an official term.
See it in action
You're at 80% of your session with two hours to go. You switch to a smaller model for quick questions, save the big rewrite for after the 3pm reset, and ask everything else in one message.
What people get wrong
- Your allowance isn't a fixed number of messages. Long chats, big files, web searches and stronger models all use it up faster.
- A long conversation costs more with every message, because the AI re-reads the whole thing each time. Starting a fresh chat often saves more than you'd think.
Go deeper
- GuideAnthropic, Claude Help CenterUsage limit best practicesWhat uses up your allowance, and how to make it go further.
- GuideAnthropic, Claude Help CenterUse Claude Code with your Pro or Max planHow chat and coding share one allowance on a paid plan.
- GuideAnthropic, Claude Platform DocsToken countingFor the curious: how tokens are actually counted.
Where it comes up
Related jargon
Drafted with AI, checked by the AI Council (Claude, ChatGPT and Gemini) and signed off by Phin.