Tokens
The basic unit an AI model reads and writes. Roughly 3-4 characters of English per token, so 750 words equals about 1,000 tokens. Every API call is priced by tokens in and tokens out.
Tokens are how AI models actually see language. Not words, not characters, but sub-word chunks chosen by the model’s tokenizer. The English word “understand” might be one token. A rare technical term might split into three. A Japanese kanji might be one. A long URL could be dozens. This matters because tokens are how the model is priced, how memory is measured, and how speed is benchmarked.
The Simple Version
Every prompt you send and every response the model returns gets counted in tokens. Providers charge per million tokens in and per million tokens out. Input and output are priced differently. Output is usually three to five times more expensive than input. The cost of running an AI feature over a year is almost entirely a function of how many tokens it consumes per user action.
Why It Matters
Token thinking is the difference between an AI feature that costs $500 a month and one that costs $50,000 a month at the same user count. Operators who ignore tokens build things that cannot scale. Operators who design for tokens know which user actions are cheap (a single prompt) and which are expensive (long tool-using agents with large context). The constraint is the design.
How It’s Used on This Site
Every cost decision on TWO is measured in tokens first, dollars second. The daily digest’s three-agent pipeline consumes roughly 50,000 tokens per run. That maps to pennies per edition, which is why we ship daily instead of weekly. Token math is the economics every operator needs to internalize.