Usage limits in Claude Code (windows, caps, and how not to hit them)

What it actually is

A usage limit is a cap on how much you can use Claude before you’re told to wait, pay more, or upgrade. In Claude Code there are two of them, stacked: a five-hour rolling session window, and a separate weekly ceiling that sits on top of it. Think of it like a phone plan with both a daily data cap and a monthly one: you can be fine on today’s cap and still get throttled because the month’s allowance is gone, or vice versa. Both clocks run independently, and both need room left in them for your next prompt to actually go through.

The five-hour window starts on your first message of a session, not on a fixed clock time, so if you start at 10am it resets at 3pm regardless of how much or little you did in between. The weekly cap resets on a fixed day and time tied to your account. Neither is measured in “number of messages.” Both are metered in tokens, the same unit that measures everything else Claude reads and writes (see tokens vs credits).

Why your AI just did this

Claude Code counts every prompt you send, every file it reads, every tool call it runs, and every line of conversation history it’s still holding in context, against the same token pool, the exact pool that also covers your regular claude.ai chat use on the same account. So a session where the agent has been reading through your codebase, running several tool calls per step, and carrying a long history forward, burns through both the five-hour window and the weekly cap noticeably faster than the same amount of time spent just typing questions into a chat window. That’s usually the real reason a heavy afternoon of agentic building empties your weekly allowance faster than a week of lighter chat use did.

When you’ll run into it

• A message mid-task reading something like “5-hour limit reached, resets at 3:42pm,” right when you wanted the agent to keep going
• A warning that appears a bit before the hard stop, easy to miss if you’re not watching the terminal
• The weekly cap running dry earlier in the week than usual, after a day or two of long agentic sessions with lots of file reads and tool calls
• On a Max plan, Opus-heavy work draining faster than Sonnet-heavy work, because Opus has its own weekly reset schedule separate from the rest
• Getting cut off mid-edit, with the agent’s current step unfinished, and the reset still hours away

What to check

• Run /usage in Claude Code before starting a big task to see the five-hour and weekly bars, and the exact time each resets
• Check whether you’re on Pro or Max: Max plans carry a higher cap, and Opus usage resets on its own weekly schedule separate from other models
• If a warning appears mid-session, wrap up and commit or save your current step rather than kicking off a new large task
• Look at whether you also used claude.ai chat earlier the same day or week; it draws from the same pool as Claude Code, it isn’t a separate allowance
• If you’re consistently hitting the cap before reset, decide between enabling usage credits (pay-as-you-go top-up on your plan), upgrading tier, or moving heavy work to an API key billed separately

Say it like a dev

Instead of: “Claude just stopped working on me” Say: “I hit my 5-hour usage limit, it resets at 3:42pm”

Instead of: “why did my week’s quota disappear so fast” Say: “that long agentic session burned a lot of tokens on file reads and tool calls, it’s the same weekly pool as my chat use”

Instead of: “just give me more, unlimited” Say: “should I enable usage credits, or move this workload to an API key so it’s pay-as-you-go instead of capped”