Claude Code cost trackers, compared — honestly
There are several good ways to track what you're spending on Claude Code, and the right one depends on what you care about. Below is an even-handed look at the main options — including ours. Where another tool is the better fit, we say so. The one thing we'd flag up front: most trackers reconstruct cost from local session logs (an estimate), while WTClaude reads the statusline — the source behind your bill — so its terminal numbers are billing-grade. That difference matters for some people and barely matters for others; the table shows where it does.
Last updated: July 2026. Spot something out of date? It's an open-source project — tell us.
Audit your team's Claude spend, free →
| WTClaude | ccusage | Claude-Code-Usage-Monitor | ccflare | Built-in /cost | |
|---|---|---|---|---|---|
| Cost basis | billing-grade (terminal Code) | estimate | estimate | estimate | estimate |
| Interface | Terminal + web dashboard | Terminal | Terminal (real-time) | Browser | In-CLI command |
| Models priced | Opus 4.8 · Sonnet 5 ($2/$10 → $3/$15 Aug 31) · Fable 5 | Log-derived | Log-derived | Log-derived | Anthropic-native |
| Cost data source | Statusline (billing-source) | Session logs (JSONL) | Session logs (JSONL) | Session logs (JSONL) | Client-side estimate |
| Real-time burn / predictions | Live monitor | — | Yes (its specialty) | Charts/filters | Session view |
| Dual-pool ready | Built for it | varies | varies | varies | Anthropic-native |
| Total across multiple machines | Yes | Local | Local | Local | Local |
| Open source | Yes (MIT) | Yes | Yes | Yes | No (built-in) |
| Price | Free | Free | Free | Free | Included |
| Runtime | Node | Node | Python | Node | Built-in |
Sonnet 5 became the default Claude Code model on June 30. A cost tracker that doesn’t price claude-sonnet-5 is mis-costing the majority of current sessions — the model most of your work now runs on. WTClaude prices it correctly ($2/$10, stepping to $3/$15 on Aug 31), off the model’s own reported tokens.
The blind spot in FinOps tools — and why ours isn't
Most cost tools read Anthropic's Claude Code Analytics API. That API only returns API-key users — your OAuth/subscription seat developers (the most common setup) never show up. So the third-party FinOps tools you'd otherwise use can be blind to your actual Claude Code developers — one org went from 9+ visible users to 1–2 after moving engineers to OAuth. WTClaude's local collector reads cost where Claude writes it, so it sees those developers.
For SMB teams: how WTClaude compares
Running Claude for a team pulls in a different comparison set — Lineman (the newer, Claude-Code-only cost tool), the FinOps incumbents (Mavvrik, Torii, Zylo, CloudEagle), and Anthropic's own native dashboards. Facts below are as of July 2026 — competitor state moves, so treat this as a snapshot, not a permanent scorecard.
| WTClaude | Lineman | FinOps incumbents | Anthropic-native | |
|---|---|---|---|---|
| Cost basis | Anthropic-computed (terminal Code) + billing-grade Cowork (Pro) | Token estimate (interceptor) | Token estimate (Analytics API) | Anthropic's own; native |
| Sees OAuth/subscription Code seats | Yes (local read) | Yes (interceptor) | No — the Analytics API returns API-key usage only | Yes (native) |
| Surfaces covered | Code + Cowork + Chat picture | Claude Code only | Multi-vendor SaaS; Claude shallow | All native surfaces |
| Independent of what it measures | Yes (sells no tokens, seats, or compression) | No (sells context compression) | Yes (but Claude-shallow) | No (is the vendor) |
| Recommendations / coaching | Yes — recommend-only, never enforce | Governance flags | Allocation / right-sizing | No — shows data, not what to do |
| Real-time | Live (under-the-cap) | ~15-min (advertised) | Daily / batch | Daily (Enterprise suite "updated daily") |
| Maturity | — | ~1-person beta — by its own disclosure, the figures shown are a synthetic demo team | Enterprise, quote-priced | Anthropic |
Competitor facts as of July 2026 — this space moves; tell us if something's out of date.
Lineman is a newer, Claude-Code-only tool whose cost view is a token-based estimate, funded off
the context-compression it sells; by its own disclosure the figures on its site are a synthetic
demo team — and it's actively shipping (@lineman-io/cli v1.0.9, Jul 1). WTClaude reads
Anthropic's own cost figure, spans Code + Cowork, and sells you nothing but the tool.
- Already living in Datadog Cloud Cost Management or CloudZero? Anthropic's Analytics API now pipes there first-party — that may be enough (it's still OAuth-seat-blind, so subscription-seat developers won't show up).
- Need multi-vendor SaaS governance beyond Claude? A FinOps incumbent (Mavvrik, Torii, Zylo, CloudEagle) is the broader fit.
- On subscription/OAuth Claude Code seats, want Anthropic-computed accuracy across Code + Cowork, and recommendations, not just a dashboard? That's WTClaude's lane — start with the free spend audit.
When to use which (the honest guide)
- Want a real-time, in-terminal burn-rate monitor with predictions? Claude-Code-Usage-Monitor is purpose-built for that and does it well.
- Want a rich browser view built from the same JSONL approach? ccflare is a nice option.
- Want the fast, ubiquitous CLI for log-based reports by day/session/project? ccusage is the one most people know, and it's good at what it does.
- Just want a quick gut-check inside a session?
The built-in
/costis right there — just remember it's an estimate. - Want billing-grade accuracy (not a reconstruction), dual-pool readiness, and one total across every machine?
That's where WTClaude is different — it reads the statusline instead of the logs. Try it:
npx wtclaude setup, thenwtclaude compareto see your own gap.
FAQ
What's the best alternative to ccusage? +
It depends what you want. For a real-time terminal monitor, Claude-Code-Usage-Monitor; for a browser dashboard, ccflare; for billing-grade accuracy from the statusline (rather than a log-based estimate) plus multi-machine totals and billing-change readiness, WTClaude. All of these are free and open source.
Are these trackers accurate? +
Most reconstruct cost from local session logs — a solid estimate, but an estimate. WTClaude reads the statusline (the source behind your bill), which is billing-grade for Claude Code in the terminal. Here's why the two can disagree.
Is WTClaude affiliated with Anthropic? +
No — independent, free, open source (MIT).
Which is ready if Claude’s billing changes? +
The June-15 Agent-SDK split was paused before it took effect. WTClaude already has the dual-pool view built, so if a revised version returns it lights up with no reinstall — meanwhile your terminal numbers are billing-grade today. More on the split.
See where your real number lands:
npx wtclaude setup →
wtclaude compare.
npx wtclaude setup