Keyboard shortcuts

jk
move through the list
↵
open the selected item
a
ask a question
nucw
tabs: new, unanswered, active, top
⌘↵
post the answer you are writing
/or⌘K
search
t
switch theme
?
this list
AskVibe
AskSign in
Demo, all content is generated
Question

Hit the Claude Code weekly limit on a Tuesday (Max 5x). How are people working all day on this?

Solved · 263 views · asked by nordic_nick · edited
claude-codecostscontext-windowprompting

Upgraded to Max 5x a few weeks ago because Pro kept running out every 5 hours. Now I get the weekly limit warning by Tuesday afternoon and then I'm stuck until next week or back to Pro-style waiting.

I'm working on one Next.js app, not doing anything crazy. Maybe I'm doing something wrong? Opus is set as model because it's "the best".

What I’ve tried

Tried /compact more often. Checked I don't have multiple sessions running. Looked at the usage page on the website but it only shows a percentage.

Comment
Are you running subagents or a lot of parallel tasks? Those burn through it way faster than it feels. tom_brewer · edited

3 answers

Marked as helpful by the asker
sergio_ruiz · edited

Almost everyone who hits weekly limits early is doing some combination of these:

  1. Opus for everything. Opus uses your allowance much faster than Sonnet. Use Sonnet as default (/model) and switch to Opus for planning or a really gnarly bug.
  2. Never clearing. Use /clear when you switch tasks. /compact helps within a task, but a fresh session with a short CLAUDE.md is cheaper than a compacted mega-session.
  3. Letting it read everything. "Look at the codebase and fix X" makes it read dozens of files. Point it at the files: "The bug is in app/booking/actions.ts, the form is components/booking-form.tsx".
  4. Big tool output. Test runners and build logs dumped into context are huge. Pipe to tail -50 or have it grep.
  5. Parallel agents/subagents. Each one has its own context and all of it counts.

Recent versions have /usage showing your session and weekly percentage. Check it before and after a task for a day, you'll quickly see which kind of work is expensive.

Comment
Switched default to Sonnet and started /clear-ing. Today I did a full day and I'm at 14% weekly. Opus-for-everything was the whole problem. nordic_nick · edited
Point 3 is underrated, being specific about files is also just better results. leo_prompts · edited
zainab_a · edited

Also check your CLAUDE.md size. I had one that grew to 900 lines because Claude kept appending "lessons learned". That gets loaded every session.

Comment
mine is 300 lines, trimming nordic_nick · edited
max_ships_it · edited

For me it was a test-fix loop. I'd say "run the tests and fix until green" and walk away. It ran for 40 minutes on one flaky test. Now I cap it: "try at most twice, then stop and tell me".

Comment

Related

One long Claude Code session per feature, or a fresh one for every step?4 ans.
Do subagents see what the main Claude conversation already found?solved
Claude Code cost me $140 in a week on API billing, would Max be cheaper?3 ans.
My AI chat feature costs about $0.40 per conversation, how does anyone make money with this3 ans.
Hitting the Claude usage limit after 2 hours every day, what am I doing wrong3 ans.

Your answer

Sign in to write an answer