A model does not read words. It reads tokens. A token is a small piece of text, sometimes a whole word, sometimes part of one. As a rule of thumb, 100 tokens is about 75 words.
Counted in Both Directions
Everything you send to the model is input: your question, your documents, the conversation so far. Everything it writes back is output. And output usually costs several times more than input.
A Worked Example
The video uses round numbers. You send a 10-page document and ask for a one-page summary. That is about 5,000 tokens in and 500 out. If input costs $3 per million tokens and output $15, the request costs about two cents.
Two cents sounds like nothing. But an agent working on a task may make hundreds of such calls, and it rereads its whole history each time. So the tokens, and the bill, grow much faster than the work seems to.
Input is ten times the output here, yet output costs several times more per token. Round numbers for illustration.
Why It Matters
This is why the token is the basic unit of the AI economy. Models are priced in tokens. Chips are measured in tokens per second. Data centres in tokens per watt.
Tokens also set a limit. A model can only hold a fixed number at once. That is its context window. Long documents and long conversations fill it up.
Tokens per Watt builds on Tokens. Tokens per watt measures a data centre’s output in tokens, the unit the tokens lesson defines.
Reasoning Models is priced by Tokens. Reasoning is made of tokens: you pay for every word of the working, even when you never see it.
Inference is priced by Tokens. Every answer uses chips and electricity, and models are priced in tokens, the unit the AI economy bills in.
Context Window builds on Tokens. Instructions, documents, conversation and the answer all share the window, and all of it is counted in tokens.
Tokens is an example of Usage-Based Pricing. Models priced per token in and out put a meter on work, the usage-based model rather than seats.
Prompt Caching(lesson unlocks soon) builds on Tokens. Caching makes the repeated part of a request cost a fraction, and the tokens lesson explains everything sent is billed as input.
Cost per Task(lesson unlocks soon) builds on Tokens. Cost per task starts from the token bill the tokens lesson defines, then adds the retries and the person fixing what the model missed.
The Business Engineer framework maps where each part of the AI stack sits and who holds the leverage. These pills are short lessons from the same library.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
