Daily burn
SpinUp Forge · Token burn
AI usage by day, source, and work driver.
Exact local logs for Codex and Claude Code, labelled estimates for local models — bucketed by London day and pointed at one question: what should the computer do next?
Data through 2026-07-28 · last extracted 2026-07-28
Claude Code turns93,714assistant turns
Active days96rows in view
Real input + output1.6Bwork floor, excl. cache
Context processed22.1Bincl. cached re-reads
Weekly trend
Log-scaled trend
Source split
Exact beside estimated
exact275.0MCodex / 1%
exact20.7BClaude Code / 94%
exact524.0KPerplexity Sonar / 0%
estimated1.1BLocal · MLX / 5%
estimated1.7KLocal · Ollama / 0%
Read Claude Code as context throughput, not spend. Only 0.7% (145.9M) is real input+output; the rest is cached-context re-reads that scale with conversation length. MLX is a prompt-token floor (output unlogged) — true usage is likely 2–4× higher.
Drivers
What is burning tokens
review46%
research37%
shipping16%
planning1%
Claude Code by estate
Foundry (harness)71%
KF (Knowledge Forge)22%
Other4%
Council3%
Scale equivalents
Make the number human
Real input + output only — excludes cached re-reads
Approx. words1.2Btokens × 0.75 words
Reading time77,800 h250 words per minute
Novel equivalents12,96790k words per novel
If every token were a written word
1.2B words
1.6B real tokens × 0.75 · excludes cached-context re-reads
That is 12,967 novels — each book below = 217 novels
≈ 1,988× the length of War and Peace
Tolstoy’s novel runs to about 587,000 words — the canonical doorstop.
1.2B words ÷ 90,000 words/novel = 12,967 novels · 2,334,001 printed pages (÷ 500 words/page) · ÷ 587,000 words (War and Peace) = 1,988×
Reading it aloud at 250 words/min
Your token burn, read aloud77,800 h
A 40-hour work-week40 h
A 2,000-hour working year2,000 h
1.2B words ÷ 250 wpm ÷ 60 = 77,800 h · 1,945 work-weeks · 39 working years · 9,725 eight-hour days
Day detail · peak day
2026-07-11
research
Total burn1.9B
Claude Code1.9B6,447 turns · 0.6% real I/O
Codex14.9M
Local · MLX22.0M41,581 req · floor
Local · Ollama0
Estate (CC)KF 92% · Foundry 8% · Other 0%
Project-level detail (commits, top projects) loads only when running locally.
Moving-average table
Last 30 days
Click a row for that day’s detail.
| Date | Total · M | 7d avg · M | Codexexact · M | Claude Codeexact · M | CC calls | Sonarexact · K | MLXest · M | Ollamaest | Driver |
|---|---|---|---|---|---|---|---|---|---|
| 2026-07-28 | 82.9 | 328.7 | 0.2 | 74.4 | 497 | 0.0 | 8.3 | 0 | review |
| 2026-07-27 | 965.9 | 428.4 | 11.6 | 922.3 | 2,938 | 0.0 | 32.0 | 0 | research |
| 2026-07-26 | 214.6 | 448.5 | 1.5 | 187.3 | 1,756 | 35.8 | 25.8 | 0 | research |
| 2026-07-25 | 114.1 | 551.7 | 0.1 | 85.4 | 610 | 0.0 | 28.5 | 0 | review |
| 2026-07-24 | 70.2 | 689.3 | 0.0 | 41.6 | 171 | 0.0 | 28.6 | 0 | planning |
| 2026-07-23 | 226.2 | 783.3 | 3.8 | 215.6 | 1,206 | 0.0 | 6.8 | 0 | shipping |
| 2026-07-22 | 626.6 | 811.2 | 7.7 | 577.1 | 2,185 | 21.8 | 41.8 | 0 | research |
| 2026-07-21 | 781.4 | 829.2 | 4.4 | 741.9 | 3,229 | 14.7 | 35.1 | 0 | research |
| 2026-07-20 | 1,106.6 | 808.6 | 16.8 | 1,060.8 | 4,514 | 0.0 | 29.1 | 0 | review |
| 2026-07-19 | 936.9 | 782.2 | 17.2 | 888.1 | 4,747 | 35.2 | 31.7 | 0 | research |
| 2026-07-18 | 1,077.2 | 725.3 | 44.9 | 1,007.5 | 4,860 | 0.0 | 24.8 | 0 | review |
| 2026-07-17 | 727.9 | 842.6 | 7.7 | 694.9 | 2,316 | 0.0 | 25.2 | 0 | review |
| 2026-07-16 | 421.9 | 841.2 | 2.6 | 393.1 | 1,622 | 0.0 | 26.2 | 0 | shipping |
| 2026-07-15 | 752.6 | 943.6 | 6.3 | 721.3 | 3,616 | 8.6 | 25.1 | 0 | shipping |
| 2026-07-14 | 636.9 | 968.2 | 10.7 | 605.5 | 2,341 | 0.0 | 20.7 | 0 | shipping |
| 2026-07-13 | 921.8 | 938.0 | 1.8 | 896.6 | 3,326 | 0.0 | 23.4 | 0 | review |
| 2026-07-12 | 538.8 | 965.3 | 1.0 | 515.0 | 2,999 | 31.6 | 22.8 | 0 | review |
| 2026-07-11 | 1,898.6 | 974.3 | 14.9 | 1,861.7 | 6,447 | 0.0 | 22.0 | 0 | research |
| 2026-07-10 | 717.7 | 841.2 | 6.4 | 691.7 | 2,762 | 0.0 | 19.6 | 0 | research |
| 2026-07-09 | 1,138.5 | 828.9 | 13.5 | 1,105.0 | 4,449 | 9.1 | 20.1 | 0 | review |
| 2026-07-08 | 925.0 | 685.1 | 5.7 | 903.1 | 3,384 | 0.0 | 16.2 | 0 | shipping |
| 2026-07-07 | 425.7 | 673.3 | 1.4 | 407.5 | 2,243 | 4.1 | 16.8 | 0 | shipping |
| 2026-07-06 | 1,113.1 | 636.4 | 9.3 | 1,085.9 | 5,030 | 68.4 | 17.8 | 0 | review |
| 2026-07-05 | 601.3 | 524.8 | 0.6 | 583.1 | 3,478 | 24.2 | 17.5 | 0 | review |
| 2026-07-04 | 967.1 | 533.4 | 10.8 | 939.0 | 4,864 | 4.1 | 17.3 | 0 | review |
| 2026-07-03 | 631.3 | 415.3 | 3.5 | 609.6 | 3,715 | 8.7 | 18.2 | 0 | research |
| 2026-07-02 | 132.0 | 363.3 | 0.1 | 117.5 | 874 | 0.0 | 14.4 | 0 | research |
| 2026-07-01 | 842.7 | 368.9 | 2.4 | 825.5 | 3,567 | 21.9 | 14.7 | 0 | research |
| 2026-06-30 | 167.2 | 265.3 | 2.6 | 150.8 | 904 | 0.0 | 13.9 | 0 | review |
| 2026-06-29 | 331.9 | 243.5 | 3.0 | 316.2 | 1,618 | 0.0 | 12.7 | 0 | review |
Method & fidelity
How each lane is counted
Days are bucketed in Europe/London. Exact lanes come from real local logs; estimated lanes are labelled and never presented as exact.
- exactCodex — sum of per-turn
last_token_usage.total_tokensfrom~/.codex/sessions/**/rollout-*.jsonl. - exactClaude Code — sum of
input + cache_creation + cache_read + outputper assistant turn from~/.claude/projects/**/*.jsonl, deduplicated by line uuid. This is context tokens processed: ~99% is cached-context re-reads, only ~1% is real input+output.CC calls= assistant turns with usage. - exactPerplexity Sonar — API-reported
prompt + completiontokens per live-search call, from the localsonar-searchCLI’s usage log. Powers weekly market-signal diffs and ad-hoc research. - estLocal · MLX — prompt tokens measured from the
mlx_lm.serverlog (progress: x/Y⇒Yprompt tokens per request). Output tokens are not logged, so this is a conservative prompt-token floor; true usage is likely 2–4× higher. - estLocal · Ollama —
≈ message characters ÷ 4from the Ollama app database; Ollama does not record token counts.