A subscription is a bet that you'll use more than you pay for. This works out what your Claude Code usage would have cost at published API rates, and divides it by what you actually pay. Paste your own numbers — nothing is uploaded, and there is no backend to upload it to.
Claude Code runs on several machines here; each pushes to the same collector.
Requires Node. Reads the local transcripts Claude Code already writes; sends nothing anywhere.
npx ccusage@latest daily --json --since $(date -d '30 days ago' +%Y%m%d)macOS/BSD: date -v-30d +%Y%m%d instead — -d is GNU-only.
Windows Git Bash uses the GNU form above.
Claude plans: Free $0 · Pro $20/mo ($17 billed annually) · Max from $100/mo with 5× or 20× usage tiers. Set the field to what you actually pay.
Same period, split by token type. Cache reads usually dominate — they are billed at a tenth of input, which is exactly why the total stays affordable.
What each day would have cost on the API. The line is your subscription's daily break-even — days above it are days the subscription paid for itself.
Token counts come from your ccusage output. They are priced with the published
per-million rates below; cache writes bill at 1.25× input (5-minute TTL) and cache reads at
0.1× input.
This is a value estimate, not a bill. On a subscription you are not billed per
token — the API-equivalent figure is what the same work would have cost had you bought it through the
API. Two honest caveats: ccusage's own total may differ from this one by a few percent,
mostly because cache-write TTL isn't always recorded and the two tools assume differently; and a
usage-tier limit is not a spend limit, so a high multiple means good value, not headroom.