Billing-grade accuracy
billing-gradeWTClaude reads the statusline data path — the same source behind your bill. Not broken logs, not estimates. Your real numbers.
Built on the right data from day one.
Most Claude Code trackers estimate your cost from session logs — which can drift from your
bill. WTClaude reads the statusline instead: the same source behind your bill. So in your
terminal the number you see is billing-grade — run
wtclaude compare
and see your own gap.
npm i -g wtclaude
wtclaude setup Free · MIT · macOS + Linux · ~60 seconds.
Your usage data is broken.
Most Claude Code trackers read JSONL session logs. Those logs record input tokens as 0 or
1 in about three-quarters of entries — they’re written mid-stream, before the response
finalizes, and never corrected. The result: the cost your tracker shows can drift from
what you actually spent — sometimes a little, sometimes a lot. A gap wide enough to choose
the wrong plan, miss a budget, or burn through usage credits without knowing. Run
wtclaude compare
to see yours.
The accuracy gap
Session-log estimate
estimate
reconstructed from the JSONL session logs
Billing-grade (your bill's source)
billing-gradeyour real cost
read from the statusline — the source behind your bill
Illustrative — your gap depends on your usage. Run wtclaude compare to see yours.
npm i -g wtclaude → wtclaude setup → wtclaude compare Built different. Built right.
Two budgets, one tracker
Built for the split (currently paused). If it returns, the dual-pool view would light up with a WTClaude update.
See where every token went
Session timelines, daily debriefs, model breakdowns, and a “what-if” plan calculator.
wtclaude compare
shows your real number next to the broken one — the moment that changes how you think
about cost.
Claude Fable tracking: WTClaude tracks Fable
5.1 and Fable 5 billing-grade in your terminal, and
wtclaude fable
forecasts what your pace would cost as usage credits on the plans that meter it (a labeled
estimate at standard API list rates; bundle discounts up to 30% and promos not reflected —
separate from the paused June-15 Agent-SDK change).
Three models now. One honest way to choose.
Opus 5.5 is the Claude Code default. Fable 5.1 is the default Fable model. Sonnet 5 has the lowest input and output rates of the three. WTClaude prices all three from your own real usage — so you choose on what it actually costs you, not the rate card.
Opus 5.5, priced right
The default in Claude Code since v2.1.280 (September 22, 2026), on every paid plan. We price it at $4/$20, with cache reads at $0.20 per million, off the model’s own reported tokens, so nothing goes untracked — and we show you the real number, not the sticker.
Fable, priced by your plan
Permanent, and it depends on your plan — not on a date. On Max, Team Premium, and Enterprise Premium it’s included up to 50% of your weekly usage limit; on Pro and Team Standard it bills as usage credits at $10/$50 from the first token. Fable 5.1 and Fable 5 share those base rates and differ on cached input — $0.25 against $1 per million tokens. WTClaude reads the billing-grade cost of each in your terminal and forecasts what your pace would cost in credits — a labeled estimate, at standard API list rates; bundle discounts up to 30% and promos not reflected.
Compare all three
wtclaude compare-models
re-prices your recorded usage across Opus 5.5, Sonnet 5, and Fable 5.1 — what would this
month cost on each? Split by surface (Code billing-grade, Cowork a labeled estimate, Chat
excluded). In the CLI and the dashboard.
Comparisons re-price your recorded usage, not the same task run on each model — every projection is a labeled estimate. We surface the cost of the choice; we don’t judge which model is “better.”
npm i -g wtclaude
wtclaude setup
then
wtclaude compare-models.
See it all on the web.
billing-gradeEverything the CLI tracks, in a dashboard you can actually look at. Your real Claude Code cost over time, a session timeline that shows where the expensive turns were, the compare view, the “what-if” plan calculator, your overall plan limit gauge with reset countdowns, and your Agent-pool forecast tile estimate .
Cost figures are billing-grade for Claude Code run in your terminal. The forecast tile is a clearly-labeled estimate.
Open your dashboard →One true total, every machine.
Work on a laptop and a desktop? Turn on opt-in cloud sync and WTClaude totals your real usage across all of them — the number no single-machine tracker can give you. (Local-first by default; nothing leaves your machine until you switch sync on, and you see a preview of what syncs before it does.)
wtclaude sync --enable Opt-in · privacy-first · preview before anything leaves your machine.
Ready when Claude’s billing changes.
Anthropic announced an Agent-SDK billing split for June 15, then paused it before it took effect. Nothing has changed — WTClaude's dual-pool view is built and ready if a revised version returns.
Anthropic announced that agent and claude -p
usage would move to a separate dollar budget — then paused it on June 15 before it took
effect. If a revised split lands, WTClaude tracks it per pool (the pool split is a heuristic),
switched on with a WTClaude update.
Daily Agent-pool forecast
estimateA projected daily and monthly agent-pool spend — and, if a revised split lands, a comparison against the credits it includes. Install early — every day you track sharpens the projection.
The cost math is billing-grade; the pool split is a heuristic, so this is a forecast — never your exact future bill.
Credits-enough check
estimateA one-line credits-enough check — ready the moment a split does land.
Install now to get yours →We tell you which numbers are real.
billing-grade inferred estimate
Every number is labeled. Billing-grade for Claude Code cost. Inferred where we have to infer. Estimate for forward-looking projections. We never dress an estimate up as exact. (Today WTClaude tracks Claude Code billing-grade in the terminal and Cowork as a labeled estimate from its audit log; Chat and a unified view are coming next.)
Free and open-source. Forever the core.
The tracker is free and MIT-licensed. What’s coming is Guardian — a paid intelligence layer (personalized recommendations, anomaly detection, spend forecasting, coaching) at $2.99/mo · $24.99/year (≈$2.08/mo — save 30%). Install the free tracker now and Guardian inherits your full history from day one.
The tracker, the dashboard, and sync are all free and open-source. Guardian (the paid intelligence layer, $2.99/mo · $24.99/year (≈$2.08/mo — save 30%)) is what’s next.
Your tracker’s been guessing. Stop guessing.
npm i -g wtclaude
wtclaude setup The accuracy story
Why Your Claude Code Cost Tracker Disagrees With Your Bill →See everything WTClaude does.
Billing-grade accuracy, multi-machine sync, a full dashboard, billing-change readiness — the most complete Claude Code tracker.
Explore all features →