If Claude Code just told you the rate limit is reached, the confusing part is that there are actually two separate limits running at the same time: a 5-hour rolling window and a longer weekly cap. You can have plenty of 5-hour budget left and still be locked out for the week — or vice versa. Knowing which one you hit is the first step to fixing it.
Fast answer
- Claude Code uses a 5-hour rolling window for burst usage and a separate weekly cap for total compute.
- In May 2026 Anthropic doubled the 5-hour limits on Pro, Max, Team, and seat-based Enterprise — but the weekly caps did not change.
- Run /usage inside Claude Code (or check claude.ai/settings/usage) to see exactly which limit you hit and when it resets.
- Most people hit limits early because of context bloat, oversized files, and leaving Opus selected for trivial tasks.
The two limits, explained
The 5-hour rolling window is burst protection. It refills on a rolling basis, so if you started a heavy session at 9am, that capacity frees up gradually through the early afternoon. The weekly cap is the real ceiling: it limits your total active compute across the whole week, and it does not refill until your weekly reset. When people say "I hit my limit on Tuesday and could not use Claude Code until the weekend," they hit the weekly cap, not the 5-hour one.
When you exceed either limit, your sessions go read-only — you can review and copy existing code, but you cannot send new prompts until the relevant window resets.
What changed in May 2026
Anthropic made two changes to Claude Code limits in May 2026: it doubled the 5-hour rate limits for every paid plan, and it removed the peak-hours throttle that used to cut available quota for Pro and Max during the busiest time slots. Both were real improvements for burst usage. The catch: the weekly compute cap was untouched. If your problem is "I run out by Wednesday," doubling the 5-hour window does not solve it — you are limited by the weekly bucket.
Why you drain limits faster than you expect
- Context bloat: long-running sessions keep re-sending the entire conversation. The longer a thread, the more every new message costs.
- Oversized file reads: pointing Claude Code at a huge directory or large files pulls thousands of tokens per turn.
- Model choice: leaving Opus selected for simple edits burns far more than routing trivial work to a cheaper model.
- Re-running instead of resuming: starting fresh sessions for the same task re-pays the context setup cost every time.
How to actually stop hitting the limit
- Check /usage first. Know whether you are limited by the 5-hour or weekly window before you change anything.
- Keep sessions tight. Start a new session per task instead of one mega-thread, and avoid loading the whole repo when you only need a few files.
- Route by difficulty. Use a cheaper model for boilerplate and reserve Opus for genuinely hard reasoning.
- Watch your burn rate live. The single biggest win is seeing how fast you are consuming the weekly cap before it runs out — not after.
- If your workflow is already clean and you still cap out weekly, that is the real signal to consider a higher tier.
FAQ
How do I see when my Claude Code limit resets?
Run the /usage command inside Claude Code, or open claude.ai/settings/usage. Both show your current 5-hour and weekly consumption plus the reset time for each window.
Did Claude Code limits get bigger in 2026?
Yes — in May 2026 the 5-hour rate limits doubled on Pro, Max, Team, and seat-based Enterprise plans, and the peak-hour throttle was removed for Pro and Max. The weekly compute cap did not change.
Why did I hit the weekly limit so fast?
The most common causes are context bloat from long sessions, loading large files or whole directories, and leaving the most expensive model selected for simple tasks. Tightening those three things usually buys back a large chunk of your weekly budget.
Should I upgrade from Pro to Max?
Only after you fix workflow waste first. If you are burning the weekly cap on bloated context and over-powered model choices, upgrading just makes an inefficient setup more expensive. If your workflow is already lean and you still cap out, Max (5x capacity) becomes easier to justify.
Deciding between Claude Pro, Max, and other plans?
Use the AI plan comparator to see Claude Code tiers alongside ChatGPT, Cursor, and Copilot before you upgrade.