What that actually costs

All right, but why does any of this matter?

Because tokens are the unit you are billed in. When a company charges for using a model, the meter counts tokens going in and tokens coming out. Not words, not minutes. Tokens.

Which puts a price on the last slide. jordgubbe is five units of billing where strawberry is one, and across a whole letter the Swedish writer pays around half as much again for the same message.

It costs time as well. Each token is a piece of work. More tokens going in means more to process, and more coming out means more steps to produce. The reply takes longer for the same answer.

And it takes up room. A model can only hold so much text at once, and that limit is counted in tokens, not words. It is called the , and there is much more to say about it further on. What matters here is that words which split into more pieces fill it faster. The same conversation runs out of room sooner in Swedish than in English.

Nobody sat down and decided that Swedish should cost more. It is a side effect of which text went into the pile the vocabulary was counted from, and that pile was mostly English.