Attilalegend

@attila

Worth at list API rates

$10,238

What these tokens would have cost billed per token at each model's published price. It is the one figure that compares across people on different plans.

List-rate estimate / day · last 90 days

on a plan$8,715

paid per token$1,523

per active day$134.71

18.32B

tokens

241.1M on a working day

75,912

requests

241.4K each

active days

of 96 tracked

day streak

still running

Activity

+96% vs previous 30 days

18.32B tokens over 76 active days · last 12 months

Averaging 241.1M per working day, with a median of 206.2M. Busiest was Sep 12, 2026 at 1.01B. Cells are shaded by what the day was worth at list rates, not by token count. Hover a day for the models and tools behind it.

Last 12 months in full on this profile, last synced Sep 21, 2026. Counters only: no prompts, no project names, no file paths.

Drift

What got reached for, month by month

Each row is one month scaled to its own 100%, so a quiet month and a loud one take the same width and only the mix changes.

Jun 26

Jul 26

Aug 26

Sep 26

claude-opus-5grok-4.6claude-sonnet-5gpt-6-astragrok-4.5deepseek-v4.1-flashclaude-opus-4-8muse-spark-1.3-contributor-freeother

Where it goes

$10,238, split four ways

Output is a fraction of the tokens but usually most of the money: it is priced at several times the rate of cache reads.

cache read

$8,151

output

$997.97

cache write

$566.39

fresh input

$522.17

Every model here has a known rate.

Cache

Cache does most of the reading

A raw token total mostly measures re-read context. The ratios below show how the work was done.

98.7%

of everything entering the context window came from cache instead of being paid for again

351.6×

tokens of context read per token written

241.4K

tokens in an average request

cache read

98% · 17.93B

cache write

1% · 95.7M

output

0% · 52.0M

fresh input

1% · 244.6M

Models

30 models, mostly claude-opus-5

claude-opus-5

11.08B · 60.5%

$6,435 estimated at list rates · 35,731 requests

grok-4.6

1.43B · 7.8%

$848.27 estimated at list rates · 9,850 requests

claude-sonnet-5

1.11B · 6%

$278.87 estimated at list rates · 3,479 requests

gpt-6-astra

1.07B · 5.8%

$1,746 estimated at list rates · 7,919 requests

grok-4.5

774.5M · 4.2%

$276.97 estimated at list rates · 4,067 requests

deepseek-v4.1-flash

689.2M · 3.8%

$14.22 estimated at list rates · 4,313 requests

claude-opus-4-8

482.4M · 2.6%

$389.75 estimated at list rates · 2,331 requests

muse-spark-1.3-contributor-free

440.6M · 2.4%

$5.60 estimated at list rates · 1,088 requests

gpt-5.6-luna

379.3M · 2.1%

$9.76 estimated at list rates · 2,367 requests

deepseek-v4-flash

303.4M · 1.7%

$4.72 estimated at list rates · 955 requests

+20 more below one percent.

Tools

7 tools, mostly Claude Code

Claude Code

12.85B · 42,373 requests · $7,282

70.1%

Grok Build

2.20B · 13,889 requests · $1,123

12%

OpenCode

1.84B · 8,842 requests · $42.51

10.1%

Codex

1.14B · 8,668 requests · $1,750

6.2%

Antigravity

219.9M · 1,757 requests · $39.46

1.2%

Muse

71.1M · 376 requests · $0.79

0.4%

OpenClaw

175.4K · 7 requests

Rhythm

Heaviest on Wednesday

Tokens by day of the week, in the timezone each device reported.

Mon: 2.30B over 11 days

Tue: 2.33B over 11 days

Wed: 3.92B over 10 days

Thu: 2.25B over 11 days

Fri: 2.48B over 11 days

Sat: 2.56B over 11 days

Sun: 2.49B over 11 days

By month

4 months tracked, busiest Aug 26.

Jun 26: 36.3M over 5 days

Jul 26: 2.77B over 22 days

Aug 26: 7.81B over 28 days

Sep 26: 7.70B over 21 days

Hour of day is not in the sync payload yet, so the rhythm here stops at the day. See docs/tokenstat-sync-cli.md for what the CLI would need to send.

Spread

Across 32 projects

Project names stay on the device. What reaches us is an opaque hash we can count but never read, so this page shows how the work was spread, never which work it was.

project 1

67.3%

project 2

12.8%

project 3

6.3%

project 4

3.2%

project 5

2.8%

project 6

2.3%

project 7

1.7%

project 8

1.2%

Ordered by size, unlabelled on purpose. The top project is 67.3% of everything.

How the counts were read

Some tools report token counts exactly; the rest are derived from what they write to disk. The chart keeps that split visible, so an inferred number never poses as a measured one.

reported by the harness

81.4%

reconstructed, high confidence

12.2%

derived from local files

6.4%