How to see your Claude Code usage limits (5-hour and weekly)

Updated October 10, 2026 · written by the maker of TokenSight, one of the apps mentioned below.

How the two windows work

On claude.ai Pro and Max plans, Claude Code usage counts against two rolling windows at once:

• 5-hour window — short-term burst limit. Starts counting with your first request and resets 5 hours later.
• Weekly window — the 7-day budget. Resets at a fixed time each week.

Each window has its own percentage and reset time, and you're stopped by whichever fills first. A heavy afternoon can hit the 5-hour limit while you still have most of your week left; a heavy week can run out the weekly limit even if no single afternoon was extreme. If you use Claude Code with an API key, there are no windows — you pay per token instead.

4 ways to check them

1. /usage inside Claude Code

The built-in way. Run /usage (also available as /cost) in any session to see where you stand. Fine for a quick check; easy to forget mid-task.

2. Your claude.ai usage page

The same limits are shown in your account's usage settings on claude.ai. Useful from your phone, less useful while you're working in the terminal.

3. A status line that shows them all the time

Claude Code pipes a JSON blob to your status line script on every refresh. For Pro and Max subscribers it includes rate_limits.five_hour and rate_limits.seven_day, each with used_percentage and resets_at. This script turns that into a compact, color-coded line — no extra tools beyond jq:

#!/bin/bash
# ~/.claude/limits-statusline.sh — 5-hour and weekly limits with time to reset.
input=$(cat)
pct()   { printf '%s' "$input" | jq -r "$1 // empty | round"; }
reset() { local t; t=$(printf '%s' "$input" | jq -r "$1 // empty"); [ -z "$t" ] && return
          local s=$(( t - $(date +%s) )); [ $s -le 0 ] && { echo now; return; }
          [ $s -ge 86400 ] && echo "$((s/86400))d$((s%86400/3600))h" || echo "$((s/3600))h$((s%3600/60))m"; }
color() { [ "$1" -ge 80 ] && printf '\033[31m' || { [ "$1" -ge 50 ] && printf '\033[33m' || printf '\033[32m'; }; }
model=$(printf '%s' "$input" | jq -r '.model.display_name // "Claude"')
line="$model"
for w in five_hour:5h seven_day:7d; do
  p=$(pct ".rate_limits.${w%%:*}.used_percentage")
  [ -z "$p" ] && continue
  r=$(reset ".rate_limits.${w%%:*}.resets_at"); [ -n "$r" ] && r=" ($r)"
  line="$line | ${w##*:} $(color "$p")$p%\033[0m$r"
done
printf '%b' "$line"

Save it, make it executable (chmod +x ~/.claude/limits-statusline.sh), and point Claude Code at it in ~/.claude/settings.json:

{
  "statusLine": { "type": "command", "command": "~/.claude/limits-statusline.sh" }
}

You'll see something like Opus 5.5 | 5h 38% (2h40m) | 7d 44% (4d2h), turning yellow at 50% and red at 80%.

4. A menu-bar app

A status line only shows while a Claude Code session is open. To keep an eye on limits all day — and get warned before you hit them — use a menu-bar app:

• CodexBar — free and open source; shows session and weekly limits for Claude, Codex and dozens of other providers by reusing your existing logins.
• SessionWatcher — paid; limits and costs for Claude Code, Codex, Cursor and more, on Mac and Windows.
• TokenSight (ours) — reads your Claude limits from the status line above, with alerts at 50 / 80 / 95 %, next to exact token usage, per-tool context attribution and Insights that point at what's burning your budget. Mac only.

For a fuller comparison, see the best Claude Code usage trackers.

Why limits run out faster than you expect

Every request re-sends the whole conversation. Prompt caching makes repeated context much cheaper, but three things quietly use up your windows:

• Long sessions. At 150K+ tokens of context, each small follow-up still re-reads all of it.
• Cache rebuilds. Step away past the cache lifetime, switch models, or edit CLAUDE.md mid-session, and the next request re-sends everything at full price. Estimate what that costs.
• Startup overhead. Big CLAUDE.md files and many MCP servers add tens of thousands of tokens before you type anything — on every request.

TokenSight's Insights page measures all three from your own session logs.