You’re on Claude Pro or Max, mid-task, and a banner appears: You've reached your usage limit. Limit resets at 14:00. You only chatted for the morning, so how did you hit it? And the time it hits keeps changing: sometimes 3 hours in, sometimes 8 hours and fine. That’s because Claude’s usage limit isn’t a simple “N messages a day” counter. As of June 2026 there are two limits running at once: a 5-hour rolling window for short bursts, and a weekly cap on top of it.
Fastest fix: open Settings > Usage (or go to claude.ai/settings/usage) and read which limit you actually hit. If the banner shows a reset time a few hours out, it’s the 5-hour window, just switch from Opus to Sonnet and wait it out. If it says the limit resets in days, you’ve hit the weekly cap, and no amount of waiting 5 hours will help. The rest of this guide explains how the meter counts and how to stop burning it so fast.
First: which limit did you hit?
The single most useful step is checking Settings > Usage, because the fix is completely different for each.
| What the banner / Usage page says | Which limit | What actually helps |
|---|---|---|
”Resets at 14:00” (a few hours out) | 5-hour rolling window | Switch to Sonnet, wait out the window, split chats |
”Resets in 2 days” / a date | Weekly cap (all models) | Stop using Opus, wait for the weekly reset, or upgrade |
| Opus blocked but Sonnet still works | Separate Opus weekly cap (Max only) | Finish the week on Sonnet 4.6; Opus returns at weekly reset |
| Hit within minutes of starting | Shared budget already spent elsewhere | Check Claude Code / Cowork / Desktop usage on the same account |
One detail that surprises people: as of June 2026, claude.ai chat, Claude Code, Claude Cowork, and the Claude Desktop app all draw from one shared usage budget on your account. A heavy Claude Code session in the morning can use up the limit you wanted for chat in the afternoon.
Common causes (why you hit so fast)
Ordered by hit rate, highest first.
1. Opus weighted far more than Sonnet
Opus 4.7 burns the limit much faster than Sonnet 4.6 for the same conversation, because each token of a heavier model costs more against your budget. On Max plans there’s even a separate, tighter weekly cap just for Opus, so you can run out of Opus while Sonnet still works.
How to spot it: Model selector showing Opus? If the task doesn’t truly need Opus (deep reasoning, complex multi-file code), switch to Sonnet 4.6.
2. Long chat replays full history every turn
Claude is stateless. By turn 30, your 50-character question still ships 50K+ tokens of input, because every prior turn is re-sent as context.
How to spot it: Are you 20+ turns deep in one chat?
3. Uploaded a large PDF / long doc
A 100-page PDF is roughly 80K input tokens. It rides along in context every turn, so 10 turns is around 800K input tokens just for that one file.
How to spot it: Big file pinned in the conversation?
4. Repeated Artifact regenerations
Every Artifact render emits the full output again. 10 edits is roughly 10x the output tokens.
How to spot it: How many Artifacts in this chat, and how many times did you regenerate them?
5. Extended Thinking always on
Thinking-mode reasoning counts as output tokens, so output can be a few times higher than normal.
How to spot it: “Thinking” / extended-reasoning toggle enabled near the model selector?
6. Multiple bursts inside the 5-hour window
The session limit is a 5-hour rolling window. A big task at noon, another at 3, another at 5 stacks three peaks into one window and almost guarantees a wall.
How to spot it: Think back over the last 5 hours of usage.
How the limit counts (simplified)
There’s no public token formula, but the relative mechanics are consistent and match Anthropic’s own framing (Opus consumes the budget faster than Sonnet, which consumes faster than Haiku):
Per-message cost ≈ (input tokens + output tokens) × model weight × thinking factor
Relative model weight (rough, Anthropic doesn't publish exact numbers):
Haiku = lightest
Sonnet = mid
Opus = heaviest (runs out first, and has its own weekly cap on Max)
Thinking:
off = baseline
on = higher (reasoning tokens count as output)
Two caps apply at once:
- 5-hour rolling window (resets ~5h after your first message in the window)
- Weekly cap (all-models; Max also has a separate Opus-only weekly cap)In plain terms: long context + a heavier model + thinking on = the fastest path to the wall.
When it resets
There are two clocks, and they reset differently. This is the part the old “just wait 5 hours” advice gets wrong.
- 5-hour rolling window — resets about 5 hours after the first message in the window, not 5 hours after you hit the wall. Usage from before you hit still counts toward that window. Hit at 9am after a 7am burst, and you may not be fully clear until well into the afternoon.
- Weekly cap — resets on a fixed day and time assigned to your account; you get your full weekly allowance each cycle. If you’ve drained the weekly cap, waiting 5 hours does nothing — you have to wait for the weekly reset (or upgrade).
Settings > Usageshows the exact next reset for both.
Note on recent changes: Anthropic doubled the 5-hour caps for Pro and Max in May 2026 and removed the old peak-hours reduction, and temporarily raised weekly limits ~50% (through mid-July 2026 as of writing). Treat any specific number you see online as a moving target and trust Settings > Usage.
Practical implications:
- “I’ll wait until tomorrow” is usually wrong for the 5-hour window — it’s hours, not a day.
- But for the weekly cap, tomorrow may genuinely not be enough; check the reset date.
- Multiple small bursts hit the 5-hour window faster than one big burst (the peak persists across the whole window).
6 effective ways to save quota
1. Split conversations
Unrelated tasks get new chats. Web bug in one chat, emails in another, product research in a third. Each new chat resets the input-token baseline instead of re-sending an unrelated history.
2. Sonnet 4.6 by default
Reserve Opus 4.7 for tasks that genuinely need deep reasoning or complex multi-file code. About 80% of daily work runs fine on Sonnet 4.6 at a fraction of the burn — and it protects your separate Opus weekly cap on Max.
3. Big files → Project Knowledge, not chat paste
Inefficient: paste the spec doc at the top of every new chat
Efficient: add it to Project Knowledge; it's retrieved on demand instead of re-sent every turn4. Use diffs, not full rewrites
Don't say: "rewrite the whole file"
Say: "at line N, replace with ...; everything else unchanged.
Reply with only the modified portion."5. Disable Extended Thinking for everyday tasks
Email drafts, outlines, and simple questions don’t need thinking. Reserve it for genuinely hard debugging or reasoning.
6. Think before sending; don’t iterate live
Inefficient: send → read → tweak → tweak... (re-bills the whole input each time)
Efficient: write the full prompt → send once → doneHow to confirm it’s fixed
- Open
Settings > Usage(orclaude.ai/settings/usage). Confirm whether the blocker was the 5-hour window or the weekly cap, and note the exact reset time shown. - If it was the 5-hour window, switch the model selector to Sonnet 4.6, then wait until the listed reset time and send a short test message. It should go through.
- If it was the weekly cap, the Usage page is the source of truth — Sonnet may keep working after Opus is blocked (Max), but a full weekly-cap block clears only at the weekly reset date.
- In Claude Code, run
/usageto see the same numbers from the terminal, since it shares the budget.
Prevention
- Treat Opus as scarce; spend it only on tasks where deep reasoning clearly pays off.
- Long projects: maintain Project Knowledge instead of pasting files into chats.
- Open a fresh chat every 20-30 turns; don’t fight to keep one alive.
- Plan high-intensity days: heaviest tokens early, small tasks later, so a morning burst doesn’t lock you out by afternoon.
- Remember the shared wallet: a heavy Claude Code or Cowork run eats the same budget as chat.
- Genuinely not enough? Consider Max (5x or 20x) or the API (per-token pricing, no 5-hour or weekly cap).
- Don’t burst the day before a deadline — it leaves no headroom for emergencies, and the weekly cap may not reset in time.
FAQ
Does waiting 5 hours always fix it?
Only if you hit the 5-hour rolling window. If you hit the weekly cap, waiting 5 hours does nothing — you wait until the weekly reset shown in Settings > Usage.
Why is Opus blocked but Sonnet still works? On Max plans there’s a separate, tighter weekly cap for Opus. Once Opus is exhausted you can keep working on Sonnet 4.6 until the Opus weekly cap resets.
Does Claude Code count against my chat limit? Yes. As of June 2026, claude.ai chat, Claude Code, Claude Cowork, and Claude Desktop all draw from one shared usage budget on the same account.
How do I see exactly how much I’ve used?
Settings > Usage on the web (claude.ai/settings/usage), or the /usage command inside Claude Code. Both show current usage and the next reset time for each limit.
Will upgrading to Max remove the limits? No plan is unlimited. Max raises both the 5-hour and weekly caps significantly (5x or 20x tiers), but the same two-clock structure still applies. For truly uncapped throughput, the API bills per token with no rolling window.
The reset time keeps moving — is that a bug? For the 5-hour window it’s expected: the window is anchored to your first message, so as you keep using Claude the “resets at” time shifts. The weekly reset is fixed to your account’s assigned day/time.
Related
- Why Claude file generation eats more quota
- Claude answers are inaccurate
- Claude long context becomes unstable
- Claude prompt best practices
Tags: #Claude #Usage limit #Debug