Reading the summary

Running cca with no arguments prints the summary. It is the view most people use most of the time, so it is worth reading closely once.

  Claude Code usage · all time · 13 sessions across 3 projects

  Tokens   input  output  cache 5m  cache 1h  cache read
          144.5K  673.5K    503.0K      6.6M       52.0M

  Model                      tokens    cost
  claude-opus-5               20.3M  $37.87
  claude-sonnet-5             20.3M  $15.97
  claude-haiku-4-5-20251001   19.4M   $6.99

  Cost    $60.82     at API list prices
  Water   ≈ 18.0 L   about 2 minutes of shower

  ─ Cost is what this usage would cost at API rates, not what you were billed;
    Claude Code on a subscription draws from your plan allowance instead.
  ─ Water assumes 0.30 mL / 1k tokens, a rough estimate. See README.

The header

Claude Code usage · all time · 13 sessions across 3 projects

The window comes first (all time, since 2026-09-14, 2026-09-01 to 2026-09-14), then the scope. A session is one Claude Code conversation; a project is one working directory.

The token row

Five classes, because each is billed at a different rate:

Class	What it is	Typical rate
input	Fresh prompt tokens	base
output	Everything the model generates	5x base
cache 5m	Cache written with a 5-minute lifetime	1.25x base
cache 1h	Cache written with a 1-hour lifetime	2x base
cache read	Cache hits	0.1x base

Cache reads dominate, but cost little

On a real corpus, cache reads are routinely 95% of all tokens and under half the cost, while 1-hour cache writes can be 4% of tokens and a third of the cost. Token share and dollar share tell different stories, which is why the view shows both.

Thinking tokens are not a separate column. They are billed as output tokens and already counted inside output; adding them would double-count. They are reported separately in JSON output.

The model table

Models ranked by spend, with abbreviated token counts. A model with no rate in the table shows unpriced instead of a figure — its tokens still count, its dollars cannot. See Rate table.

Cost and water

Cost    $60.82     at API list prices
Water   ≈ 18.0 L   about 2 minutes of shower

Cost is the sum of every priced token at published API rates.

Water is a deliberately rough estimate, printed with its assumption inline every single time. It is a placeholder, not a measurement — see The water estimate.

The footnotes

Two always appear:

• that the cost is not a bill, because a subscription draws from an allowance;
• that the water figure rests on a rough assumed rate.

Others appear only when they apply — unknown models, missing cache tier data, fast mode, US-pinned inference, batch tier, web search, undated records. Each names a condition that changed the arithmetic. If you see one, it is telling you something specific about your data.

Colour

Output is coloured when writing to a terminal that supports it:

• orange for token counts
• green for money
• light blue for the water figure
• amber for anything you are being warned about

Colour is disabled automatically when output is piped or redirected, and can be turned off with --no-color or the NO_COLOR environment variable.