Token Quota

AI systems that write text break down every input into small pieces. Such pieces are called tokens. A token is usually a short word or part of a word, often just two to four letters in German. The provider of such a service does not bill by sentences or pages, but by exactly these pieces. A token quota is the amount of pieces one is allowed to consume before it stops or becomes more expensive. The quota can apply per minute, per day, or per month.

Why every word costs money

Behind every answer from an AI system are specialized chips that consume a lot of electricity. The longer the input and the longer the answer, the more computing time is incurred. A provider cannot cover these costs with a flat rate, because usage fluctuates strongly. That is why it measures consumption in tokens and caps it.

For companies, the quota is an important planning figure. Anyone building software that makes an AI request for each customer must first calculate how many tokens that will add up to per month. If one miscalculates, the bill can turn out to be many times higher than planned. Some companies therefore set themselves a hard limit so that a faulty program does not burn through thousands of euros overnight.

The topic is also noticeable for private users. Free access to chatbots almost always comes with a hidden quota. Once it is used up, the system switches to a weaker model or requires a waiting period. The limit is therefore not a technical coincidence but a deliberate commercial decision.

How it is counted and capped

Two directions are counted separately. Input tokens are everything that is sent in: the question, attached files, and often also the previous conversation history. Output tokens are what the model writes back. Output tokens usually cost significantly more than input tokens, often three to five times as much.

It is important to distinguish this from the context window. The context window is a model’s working memory for a single request, i.e. how much text it can take in at once. The quota, on the other hand, is a sum across many requests. A model with a huge context window is of no use if the monthly quota is small.

Technically, the quota is usually regulated via two counters. One limits tokens per minute and protects the servers from overload. The other limits tokens per month and protects the budget. When a limit is reached, the service responds with an error message instead of text. Good programs then wait briefly and try again instead of crashing.

From chatbot subscription to quarterly figures

The term is most commonly encountered in the price lists of AI providers. There, one finds figures such as a certain price per one million tokens. One million tokens sounds like a lot, but corresponds to only around 700,000 words, roughly five thick novels. Anyone who has an AI summarize long documents every day will quickly burn through that.

In business news, the figure appears as a measure of success. Major providers now report how many trillions of tokens per month run through their data centers. This figure is considered an indicator of how heavily AI is actually being used, similar to user numbers for social networks. As it grows, companies need more chips and more electricity, which in turn moves the share prices of chip manufacturers.

A common misconception is that one token equals one word. For German texts with long compound words, consumption is noticeably higher than for English ones. A German text can require around a third more tokens to express the same statement. Anyone calculating their quota should factor that in.