What it means
Models do not read whole words. They split text into tokens, which are common chunks of characters. In English, a token is roughly three quarters of a word on average, though it varies by language and content.
Both what you send and what the model writes are counted in tokens, which decide cost and whether a request fits in the context window.
Why it matters for a business
Token counts drive AI costs. Sending whole documents when a summary would do, or asking for long answers when short ones are enough, multiplies spending at scale.
A business example
Things to watch
-
Measure average tokens per request in production.
-
Trim inputs to what the task needs.
-
Set sensible output length limits.
-
Remember different languages use different token counts.