THE TOKEN COST CARD · what the workload costs, from your own sheet
WAKES WHEN someone asks what a model workload will cost, how many tokens, or whether caching pays
THE LOOP
tokencost.py new · copy prices in, dated
tokens · the prompt and the documents
estimate · a call, the run, cached or not
usage · the real log: by model, latency
cut the prompt before cutting the model
NEVER
Type a price the provider did not publish
Quote a figure without the sheet's date
Call an estimate a bill
IT MAKES new — the price sheet to fill in, per million tokens, dated; tokens — a rough token count (about four characters a token); estimate — a call, the run, the caching saving; usage — the log: totals, cost by model, latency p50/p95, the costliest calls.
THE STANDARD the card types no price; a figure without a date is a rumour; output tokens cost most; the system prompt is paid for on every call; measure, then estimate; an estimate is a budget line, not a bill.
PROOF SELFTEST 13/13 · witnessed in Claude 2026-10-09 · Edition 1.
THE SEAL 98 of 100 · LEGENDARY · WITNESSED · Trigger 5 · Machinery 5 · Law 5 · Portability 5 · Proof 5
IT CAN BE WRONG an AI reading a text file: it can be wrong and can invent a number, a rule, a name or a date; check what matters against your own records and the source; the risk is yours. THE-CLAUSES.md rides in the packet and is part of the licence.
THE PUNCH · SHA-256 OF SKILL.MD · 8989FC80124CF56A8C5D22F3
CITE THIS CARDTHE TOKEN COST CARD, Edition 1. Lund Studio LLC, Westminster, Colorado. https://lundr.tech/?view=token-cost-card. Punch 8989FC80124CF56A8C5D22F3. Retrieved 2026-10-10.
What it is.
Works out what an AI model workload costs and how long it takes from the team's own price sheet and usage logs: a token estimate for any prompt or document, the cost of a call and of a run of calls with the saving from cached input, and a reading of a usage log by model with latency p50 and p95 and the costliest calls — with the price sheet's date on every figure, and no price typed by the card.
It carries a machine that shows its working: new — the price sheet to fill in, per million tokens, dated; tokens — a rough token count (about four characters a token); estimate — a call, the run, the caching saving; usage — the log: totals, cost by model, latency p50/p95, the costliest calls — 13 checks in its selftest, every command on the record. It will never type a price the provider did not publish; never quote a figure without the sheet's date; never call an estimate a bill.
Six laws, written into the card.
They are in the card itself, so the AI follows them.
THE CARD TYPES NO PRICEThe figures come from the provider's page, copied by you, with the date; the sheet refuses to run on placeholders.
A FIGURE WITHOUT A DATE IS A RUMOUREvery estimate prints the sheet's date beside it.
OUTPUT TOKENS COST MOSTLength is the first lever; the model is the last.
THE SYSTEM PROMPT IS PAID FOR ON EVERY CALLCache it where allowed, trim it where it is not earning its place.
MEASURE, THEN ESTIMATEA usage log beats any forecast; the estimate is for what has not run yet.
AN ESTIMATE IS A BUDGET LINE, NOT A BILLSay so when it is quoted.
What rides in the card.
In the card’s own voice, with a machine that shows its working.
THE GUIDEreferences/WHERE-THE-MONEY-GOES.md
THE MACHINEtokencost.py · four commandsnew — the price sheet to fill in, per million tokens, dated; tokens — a rough token count (about four characters a token); estimate — a call, the run, the caching saving; usage — the log: totals, cost by model, latency p50/p95, the costliest calls. Every command prints its working and refuses a bad input; 13 checks in its selftest, every command on the record.
What it does once your AI has it.
new The price sheet to fill in, per million tokens, dated
tokens A rough token count (about four characters a token)
estimate A call, the run, the caching saving
usage The log: totals, cost by model, latency p50/p95, the costliest calls
On the record The price sheet written with placeholders and refused until filled; two files counted in tokens; ten thousand calls estimated on the example sheet — half the input cached took 32.00 to 26.60; and a forty-call usage log read by model, with p50 and p95 latency and the costliest calls named.
What it will never do.
It will never type a price the provider did not publish; never quote a figure without the sheet's date; never call an estimate a bill.
The token count from text is an estimate; real tokenizers differ by language and content, so measure with the provider's counter where it matters. A cost is only as current as the sheet's date. Nothing here forecasts what a provider will bill. It is an AI reading a text file: it can be wrong, so check what matters against your own records and the source.
What’s in the buy.
CARD.md — the whole card: who it is, its laws, when it says no, what it carries · SKILL.md — the door your AI reads first · START-HERE.md — the first five minutes, for you
WHERE-THE-MONEY-GOES.md — the guide in references/, in the card's own voice
scripts/tokencost.py — the machine: new · tokens · estimate · usage, standard library only, selftest 13/13
WITNESS.md + SELFTEST.md — every command on the record, as it printed · MANIFEST.md, every file hashed · AUDIT.md, the twelve checks
MY-LEDGER.md — the card's memory, on your own machine · TIN.md · LICENSE-NOTE.md · THE-CLAUSES.md · HOSTS.md
card.svg + card-machine.svg + card.json · card.html, the card as one offline file · the doors for ChatGPT, Gemini, Cursor, Claude Code and MCP hosts (rules/, system/, mcp/)
PROOF · SELFTEST 13/13 · 4 COMMANDS ON THE RECORD · WITNESSED IN CLAUDE 2026-10-09 · EDITION 1 · PUNCH 8989FC80124CF56A8C5D22F3
ON EVERY DOOR · A CARD FOR YOUR AI, A FILE FOR YOUR COMPUTER
In Claude: turn on code execution under Settings › Capabilities, then Customize › Skills › + › Create skill, and upload the zip as it is — START-HERE.md in the box walks you through it, and tells you how to keep the ledger. In ChatGPT, Gemini and the rest: a Project, a GPT or a Gem with the card’s files attached and the short door in the box pasted as its instructions. The words run anywhere an AI reads Markdown; the machines run wherever the host runs Python. Witnessed in Claude only. The host marks →
Take it.
Nº 836 PROPOSED — LAUNCHING SOON
Sold as is under the term printed. A card instructs an AI you operate; results depend on your model, your files and your judgement. Not medical, legal, tax or financial advice, never the person’s boss, and not a licence to practise. The terms →
One payment. The file is yours to keep and keeps working at its edition; new editions issued in the year after you buy reach you by email as replacement files. Tax is added at checkout where it applies. Nothing auto-renews.
A TERM, NOT A SUBSCRIPTION: NOTHING RENEWS ITSELF AND NOBODY IS CHARGED AGAIN. WHEN A TERM ENDS THE FILES STAY ON YOUR COMPUTER AND STAY READABLE — FREEZE, NOT LOSS.
THE MACHINE · IN THE OPEN · FROM THE WITNESSED RUN OF 2026-10-09
tokencost.py, every command showing its working.
A small Python program, standard library only. Your AI runs it; you can too. Every block below is pasted from the card’s own WITNESS.md, as printed.
new
the price sheet to fill in, per million tokens, dated
tokens
a rough token count (about four characters a token)
estimate
a call, the run, the caching saving
usage
the log: totals, cost by model, latency p50/p95, the costliest calls
selftest
13 checks that prove the machine; it must print 13/13
new · THE PRICE SHEET TO FILL IN, PER MILLION TOKENS, DATED
$ python3 scripts/tokencost.py new prices.json
price sheet written · prices.json · fill input, output and cached_input per million tokens for each model, name the currency, and write the price page's date in as_of
(the sheet filled with EXAMPLE figures for this demonstration — not any provider's prices — and dated)
$ (the sheet filled with EXAMPLE figures for this demonstration — not any provider's prices — and dated)
tokens · A ROUGH TOKEN COUNT (ABOUT FOUR CHARACTERS A TOKEN)
$ python3 scripts/tokencost.py tokens system-prompt.txt faq.md
system-prompt.txt 2,160 chars · 360 words · about 540 tokens
faq.md 1,931 chars · 277 words · about 483 tokens
TOTAL · about 1,023 tokens
NOTE · an estimate at four characters a token; measure with the provider's counter where it matters
estimate · A CALL, THE RUN, THE CACHING SAVING
$ python3 scripts/tokencost.py estimate --prices prices.json --model example-small --in 1200 --out 400 --calls 10000 --cached 0.5
ESTIMATE · example-small · 1,200 in / 400 out a call · 10,000 calls · prices as of 2026-10-09
a call USD 0.0027
total USD 26.60 for 16,000,000 tokens
caching 50% of input from cache saves USD 5.40 (would be USD 32.00)
NOTE · a budget line from your own sheet, not a forecast of a bill
usage · THE LOG: TOTALS, COST BY MODEL, LATENCY P50/P95, THE COSTLIEST CALLS
$ python3 scripts/tokencost.py usage --prices prices.json usage.jsonl
USAGE · usage.jsonl · 40 calls · 67,133 input tokens (20,000 cached) · 16,549 output · USD 0.55 · prices as of 2026-10-09
example-large 4 calls · USD 0.46 · USD 0.12 a call · 24,000 in · 4,800 out
example-small 36 calls · USD 0.09 · USD 0.0024 a call · 43,133 in · 11,749 out
latency p50 1,120 ms · p95 5,200 ms · max 5,200 ms
costliest calls:
2026-10-09T10:00:00 · example-large · in 6,000 out 1,200 · USD 0.12 · report
2026-10-09T10:10:00 · example-large · in 6,000 out 1,200 · USD 0.12 · report
2026-10-09T10:20:00 · example-large · in 6,000 out 1,200 · USD 0.12 · report
2026-10-09T10:30:00 · example-large · in 6,000 out 1,200 · USD 0.12 · report
2026-10-09T10:22:00 · example-small · in 1,446 out 449 · USD 0.0032 · chat
THE NUMBERS, NAMES AND DATES IN THESE RUNS ARE THE STUDIO’S OWN EXAMPLES, INVENTED FOR THE RECORD; NO REAL COMPANY, PERSON OR PROPERTY IS NAMED.
QUESTIONS AT THE TABLE
Asked before the wire.
What does it need from me?
Tokencost.py new · copy prices in, dated. It asks first for the provider's price page (to copy the figures, with the date), the prompts and documents in play, and the usage log if there is one.
Does it send anything anywhere?
No. The machine is a standard-library Python program that runs on your own computer and sends nothing anywhere; the words run wherever an AI reads Markdown. No server, no account, no key.
What is on the record?
The self-test (13/13) and every command run on real example inputs, pasted into WITNESS.md as they printed; the punch on the face is the SHA-256 of SKILL.md, so a downloaded copy can be checked against this room.
Can it be wrong?
Yes — it is an AI reading a text file. The token count from text is an estimate; real tokenizers differ by language and content, so measure with the provider's counter where it matters. A cost is only as current as the sheet's date. Nothing here forecasts what a provider will bill.
What will it never do?
It will never type a price the provider did not publish; never quote a figure without the sheet's date; never call an estimate a bill.
