How long does a Claude Code rate limit last?

Claude Code uses a 5-hour rolling window. When you hit the limit, you must wait until the window rolls forward — typically 1 to 5 hours depending on when you started. Heavy parallel-agent users on Pro can lock themselves out in 30 minutes.

How do I avoid Claude Code rate limits?

Three options: (1) upgrade to Max 5× or 20×, (2) batch your agent runs so the 5-hour window starts at a strategic time, (3) track real usage with SessionWatcher so you know when to slow down before lockout hits.

Does running parallel agents make me hit limits faster?

Yes, dramatically. Each parallel agent burns its own token budget against the same 5-hour cap. Two parallel agents at full speed will exhaust a Pro plan in under 20 minutes. The Lockout Risk Score above accounts for this.

Claude Rate Limit Calculator

Get a gamified 0-100 lockout risk score. Above 90 means you're minutes from a rate-limit error.

1Your plan

2Hours into your 5h window2h

3How are you using it?

Lockout Risk Score

100

About to lock

You'll likely hit lockout in 0 minutes, before your 5h window resets.

Score too high?

Find a plan with more headroom

Want exact numbers?

Try the token estimator

How the score works

We multiply your usage intensity by hours active and divide by your plan's published 5-hour token budget. Anything above 90 means you're minutes from a rate-limit error. The only way to ground-truth this is to measure your actual burn — which is exactly what SessionWatcher does.

What “lockout” actually means

Anthropic, OpenAI, and Cursor all use rolling token windows (5 hours is the common figure). Hit the cap and your next request returns a 429 error. You're not banned — you just can't use the tool until the window rolls forward. For a Claude Code user mid-session, that's a productivity disaster.

The honest limit of this calculator

We estimate based on published token caps and average burn rates per usage type. Your actual burn depends on prompt size, files in context, model choice, and tool call volume — none of which we can see. The only ground truth is a live token meter.

The fix

SessionWatcher puts your live token burn, lockout countdown, and 5-hour window progress in your macOS menu bar. $2.99 one-time. No subscription, no telemetry, just a number that stops you from hitting 429 mid-deploy.