Token Plan
A token plan is a pricing model for AI services in which you don't pay for time or the number of requests, but for the amount of text units processed. These units are called tokens, and the longer the question and answer, the more of them you consume.
When a computer program processes text, it first breaks it down into small pieces. Such a piece is called a token: usually a short word, a syllable, or a punctuation mark. As a rough rule of thumb, 100 words of English text amount to roughly 130 to 170 tokens. A token plan is a pricing model in which you pay for exactly these pieces or are allocated a fixed allowance of them. So you’re not buying usage time or a number of questions, but a quantity of text. Providers of chat and coding assistants today almost all bill according to this principle.
Why billing is based on text volume
For the provider, every answer costs real money. Specialized computing chips have to run for it, and their power consumption is almost directly tied to the amount of text. A one-sentence question is cheap, evaluating an 80-page contract is expensive. A token-based rate reflects these costs much better than a flat monthly fee.
For users, this means: your own costs are predictable, but not always intuitive. Two people with the same subscription can have completely different bills. Someone who only asks short questions can go a long way. Someone who sends entire databases through the model quickly hits the limit.
That’s why token numbers have become part of business news. When a provider lowers the price per million tokens, it’s newsworthy. In recent years, these prices for comparable performance have repeatedly dropped significantly, sometimes by more than 90 percent within a few months.
How consumption is composed
Counting happens in both directions. Everything you input is input tokens. Everything the model writes is output tokens. Output is almost always more expensive than input, often by a factor of three to five, because the model has to compute each word individually and anew.
A common misconception is that only the last question counts. In a longer conversation, the program usually sends along the entire previous history so that the model understands the context. So consumption per message increases over the course of a chat. Anyone who attaches a document pays for it again with every further follow-up question.
This should be distinguished from the context window. It describes how many tokens a model can even oversee at once, and is a technical limit. The token plan, on the other hand, is a commercial limit. A model can have a very large context window while one’s own plan is nevertheless exhausted after just a few long requests.
Where token plans show up in everyday life
They are most visible in interfaces for developers. There, a price is listed like “per one million tokens,” and billing occurs monthly based on actual consumption. Many companies additionally set upper limits so that a faulty program doesn’t burn through thousands of euros overnight.
In consumer subscriptions, the model is often hidden. Instead of tokens, people then talk about messages per hour, credits, or a fair-use limit. Behind this, as a rule, there is still a token budget. The notice that an assistant switches to a weaker model after intensive use is usually also related to this.
For everyday school use, this means quite practically: summarizing a long text consumes a lot, a short follow-up dialogue consumes little. Anyone who wants to stay economical only inputs the sections that are truly necessary and starts a new conversation for a new topic.
