Discord Download

ccusage vs AutoRouter

AutoRouter lowers what Claude Code and Codex cost you, and shows each turn’s model and cost as it happens. It sends each turn to the right model, using your plans first and then credits at the provider’s price plus 10%. ccusage is a free, open-source CLI by ryoppippi that reads your coding agents’ local logs and reports what you already spent, by day, week, month, session or 5-hour block. This page covers how ccusage works, then how the two compare. Learn more

AutoRouter A Mac app that picks a model for each turn and shows what it cost. $0/mo on your plans Download ccusage An open-source CLI for usage and cost reports. $0open source
Price $0 on your plans, Pro $50 a month Free, MIT licensed
Fee on tokens Provider’s price plus 10% on paid models None, it only reads logs
Runs on Your Mac, macOS 14 or later Your terminal, through npx, bunx or Nix
What it does Sends each turn to the right model across your plans Reports token usage and cost from local logs
Covers Codex, Claude Code, Cursor, Claude Desktop, T3 and Terminal 20 coding agents, among them Claude Code, Codex, OpenCode, Gemini CLI and Copilot CLI
Setup Install, sign in, click Enable Run npx ccusage@latest, no sign-in
Your plans Used first, then credits; moves to another account when one runs out Unchanged; costs shown at API prices
Model choice Your rules, then a model that clears the quality bar Unchanged, it routes nothing
What you see Every turn’s model, cost and why, as it runs Estimated cost by day, week, month, session or 5-hour block
Requests go Through AutoRouter’s servers to each provider Straight to the provider, as before; ccusage reads logs on your machine
  • Lower spend, with each turn sent to the right model
  • Each turn’s model, cost and why, as it happens
  • Models included, with no keys to manage
  • Rollout to every Mac in a team through MDM

Sources checked November 18, 2026: ccusage README (version 20.0.51), the ccusage docs and AutoRouter pricing. AutoRouter wrote this page. See every comparison.

How ccusage works

What the project is, how it counts cost, and how setup works.

Reports from the logs you already have

ccusage is an open-source project by ryoppippi on GitHub, MIT licensed, with more than 19,000 stars. It reads the usage data coding agents already write to disk, such as Claude Code’s files under ~/.claude and ~/.config/claude/projects, and turns it into daily, weekly, monthly and session reports with tokens, cache tokens, models and cost in US dollars (README).

It covers 20 agents, from Claude Code and Codex to OpenCode, Amp, Gemini CLI and Copilot CLI. ccusage blocks groups Claude Code usage into 5-hour windows and shows the active window’s burn rate and projected total (docs).

How it counts cost

By default, ccusage uses the cost Claude Code writes into its logs when an entry has one, and otherwise works it out from token counts with pricing from LiteLLM’s database and the models.dev catalog (cost modes). So the figures are estimates at API prices. Its docs say the analysis runs entirely on your machine and usage data never leaves it, and --offline uses cached pricing instead of fetching it.

How setup works

There’s nothing to install: run npx ccusage@latest, or use bunx, pnpm dlx or Nix. Add --since and --until for a date range, --breakdown for cost by model, or --json for scripts. For a running view, add ccusage statusline to Claude Code’s status line in settings.json; it’s in beta and shows the session’s cost, today’s total and the current 5-hour block (status line).

What it does well

  • It’s free and open source under the MIT license.
  • One npx command runs it, with no account or sign-in.
  • It reads logs on your machine, and works offline with cached pricing.
  • It’s good for auditing history: any day, week, month, session or Claude Code project, with cost by model.
  • JSON output feeds scripts, dashboards and community tools.
  • It reports 20 coding agents in one table.

Where it gets harder

  • It reports spend after it happens but can’t lower it. Every turn still runs on the model you picked.
  • On a Claude or ChatGPT plan, its dollar figures show what the tokens would cost at API prices, not what the plan charges or how much of it is left.
  • It shows which models ran, but not why, or whether a cheaper one would have done.
  • The live blocks monitor was removed in version 18. The docs point to the beta status line instead.

How AutoRouter differs

ccusage tells you what you already spent. AutoRouter lowers what you spend, and shows each turn’s cost as it happens.

Each turn sent to the right model

AutoRouter shows up as one more model in the picker you already use. While it’s selected, it works out how hard each turn is and only considers models that clear the quality bar for it. Hard work goes to the strongest model your plans cover, and quick work to the one that costs you least, often a low-cost open model, so the plan isn’t touched. Your own plain-English rules go first, such as “Use Claude Opus 5.5 for UI work.”

Plans first, then credits

Turns run on the Claude and ChatGPT plans you already pay for first. You can connect Claude, ChatGPT, Grok and Gemini plans, and when one account reaches its limit, AutoRouter moves to the one with the most room left before it spends any credit. Paid models bill at the provider’s price plus 10%.

Cost as it happens

Every turn shows its model, what it cost and why, while you work rather than in a report afterwards. You can see which models do best on which tasks and how often each rule is used, then set a rule or pick a model yourself. Enterprise adds spend by team, project or ticket. It covers far fewer tools than ccusage’s 20 agents.

Built for teams to roll out

Enterprise rolls AutoRouter out to every Mac through Jamf or another MDM, with nothing for anyone to set up. Admins get allowed models, team budgets and zero data retention in one place, spend by team, project or ticket, and SSO, SCIM and an audit log.

What it costs and what it can’t do

Routing across your plans is free. Pro is $50 a month with $20 of included usage, and paid models bill at the provider’s price plus 10%. It runs only on macOS 14 or later and isn’t open source. It sometimes misjudges how hard a task is, and it only shows turns that run through it, so it can’t report on past sessions. Requests pass through AutoRouter’s servers on their way to each provider; see privacy for what is kept.

Which one to choose

Choose ccusage if you want to audit what past sessions cost, want a free open-source tool that works offline, or use agents AutoRouter doesn’t connect, such as Amp, Gemini CLI or Copilot CLI.

Choose AutoRouter if you’re on a Mac and want to spend less, with each turn sent to the right model and its cost shown as it runs. You can also run both, with ccusage reporting on your logs while AutoRouter picks the model for each turn. For other options, read Model routers for Claude Code and Codex, compared.

Setup, side by side

The steps each one’s own docs give, as of November 18, 2026.

ccusage

  1. Run it npx ccusage@latest, or bunx ccusage, pnpm dlx ccusage or Nix.
  2. Pick a report daily, weekly, monthly, session or blocks, for every agent or one.
  3. Filter it Add --since, --until, --breakdown or --json.
  4. Add the status line Optionally, set ccusage statusline as Claude Code’s status line command.

AutoRouter

  1. Download AutoRouter for macOS, signed and notarized by Apple.
  2. Sign in Sign in or create a free account, then sign in to your plans.
  3. Enable your apps Allow Accessibility and click Enable for each app.
  4. Pick AutoRouter In a new session, choose AutoRouter from Codex’s model menu, or with /model in Claude Code.

Questions

What people ask about ccusage.

What is ccusage?

A free, open-source CLI by ryoppippi that reads the local logs of Claude Code, Codex and other coding agents and reports token usage and estimated cost by day, week, month, session or 5-hour block. Its docs are at ccusage.com.

Is ccusage free?

Yes. It’s MIT licensed, and npx ccusage@latest runs it without a global install.

How accurate are ccusage’s costs?

They’re estimates. It uses the cost Claude Code logs when an entry has one, and otherwise calculates from token counts with LiteLLM and models.dev pricing. On a subscription plan, the figures show what the tokens would cost at API prices, not what you pay.

Does ccusage send my data anywhere?

Its docs say the analysis runs entirely on your machine and usage data never leaves it. With --offline, it uses cached pricing instead of fetching it.

Does ccusage change which model runs?

No. It reads logs after the turns have run, so your agents use the model you picked. AutoRouter picks a model for each turn after your own rules, and shows what it cost as it runs.

What does ccusage blocks show?

Claude Code usage grouped into 5-hour windows, with tokens and cost for each and the active window’s burn rate and projected total. --token-limit warns as you near a limit you set. The live monitor was removed in version 18; the docs suggest the status line instead.

Can I use ccusage and AutoRouter together?

Yes. They do different jobs. ccusage reports on the logs your agents keep, and AutoRouter picks the model for each turn, showing each turn’s model, cost and why.

How is AutoRouter different from ccusage?

ccusage reports what you already spent. AutoRouter lowers what you spend: it sends each turn to the right model, uses your Claude and ChatGPT plans before any credit, comes with models included and needs no keys.

What does AutoRouter cost?

Routing across your ChatGPT and Claude plans is free. Pro is $50 a month with $20 of usage, and paid models bill at the provider’s price plus 10%. See pricing, or estimate a month with and without AutoRouter in the savings calculator.