How to stop hitting the usage limit by Wednesday
You pay for the plan, and on Wednesday it tells you to come back on Saturday. The limit is not a bug. It is a budget, and most people spend it on the wrong things.
I keep hitting usage limits. How Claude's five-hour and weekly limits and ChatGPT's message caps actually work as of October 2026, where to see your usage, and the five habits that make a Pro plan last the week: smaller models, projects, batching, fresh chats and knowing what counts.
Find out where your usage actually goes
Claude shows you. Open Settings, then Usage. The page is headed Plan usage limits and has two bars: Current session, which is your five-hour window and the time left in it, and Weekly limits, which shows when the week resets, for all models, and separately for the heaviest model if your plan includes one. In Claude Code the same numbers come from typing /usage.
ChatGPT does not have a usage page like this; it tells you when you reach a cap for a model and when it resets, and the cap depends on the plan and the model. So on ChatGPT the method is the same but you fly blind, which makes the habits below matter more.
Learn what counts
Three things decide how fast the bar moves, and Anthropic lists them on its usage page: model choice, effort level and current conversation length. The heaviest model uses the weekly budget several times faster than the standard one. A long chat costs more with every message, because the whole conversation is sent again each time. And files you paste in cost every time you paste them.
So the person who hits the limit on Wednesday is usually using the heavy model, in one long chat, pasting the same document in repeatedly. Each of those has a fix.
One chat that has run since Monday.
Pastes the brief in again each morning.
Asks one small question per message, twenty times.
Comes back after lunch and the cache has expired.
A new chat for each task, inside a project.
The brief lives in Project knowledge, pasted never.
Related questions batched into one message.
Works in blocks so the cached context is still warm.
Move everything you reuse into a project
Anthropic's own advice for using less is "use projects for anything you'll reference multiple times". The reason is caching: content in a project is cached, so the model does not re-read it in full on every message. Content you paste into a chat is read in full every time. The same brief, pasted five times, costs five times.
Make a project per piece of work, put the brief and the reference files in it, and start every chat there. It also fixes the lost-the-plot problem, which we cover in its own guide.
Batch, and be specific
Anthropic's list of tips reads like an editor's: combine related questions into one message, give clear detailed instructions, review your message before sending so you need fewer follow-ups, and refer back to earlier information instead of repeating it. Each round trip resends the conversation, so five small questions cost far more than one message with five numbered questions.
The prompt below is my template. It is boring, and it roughly halves the number of messages I send.
I have five related questions about the document in this project. Answer all five in one reply, numbered, each under 100 words. Do not repeat the question back. 1. … 2. … 3. … 4. … 5. … If any answer needs something I have not given you, say so in one line instead of guessing.
Work in blocks and start fresh often
Anthropic's page has one detail most people miss: caches expire after a period of inactivity, and when you come back after a long break the content counts in full again. So an hour of focused work costs less than the same work spread over a day in ten-minute visits.
And when a chat gets long, hand off and start a new one. Every message in a long chat resends the whole history. A fresh chat with a 300-word handoff note is cheaper and gets better answers. In Claude Code, /compact does the summary for you.
TipBefore you upgrade to a bigger plan, run a week with the standard model as default and projects for your files. Most people who do this stop hitting the limit without paying more.
You're done when
Questions people are asking
How do Claude usage limits work? There is a rolling five-hour session limit and a weekly limit, shown as bars under Settings, Usage. The weekly limit resets on a fixed day. Plans that include the heaviest model show a separate weekly bar for it.
Does Opus use more of my Claude limit than Sonnet? Yes, several times more per message. Model choice is one of the three factors Anthropic names, with effort level and conversation length.
How do I use less ChatGPT usage? Use Instant or Medium by default, keep the heavy reasoning levels for hard tasks, put reference files in a Project, and ask several questions in one message instead of one at a time.
Sources
- Claude Help Center, Usage limit best practices, Sep 22 2026 support.claude.com/en/articles/9797557-usage-limit-best-practices
- Anthropic, Maximizing the value of your Claude Code sessions, Aug 14 2026 claude.com/blog/maximizing-the-value-of-your-claude-code-sessions
- OpenAI Help Center, About ChatGPT Pro tiers help.openai.com/en/articles/9793128-about-chatgpt-pro-tiers
How to tell when an AI answer is made up, in under two minutes
One email a week. Written by people.
New how-tos, the tools we tested and what we think, in one short email.