Tokens (in AI)
Tokens are the small chunks of text an AI model reads and writes. A model doesn’t see words or letters. It breaks text into tokens, which are often whole short words, pieces of longer words, punctuation or spaces. In English, a token is about three quarters of a word, so 1,000 tokens is roughly 750 words.
Why it matters
Tokens are the unit almost everything in AI is measured in. A model’s context window, its speed and its price are all counted in them. Providers charge per million tokens, with separate rates for input (what you send) and output (what the model writes), and output usually costs more.
Token use grows in ways people don’t expect. In a long chat, every new message sends the whole conversation again, so you pay for the earlier turns again too. An agent working on a task may call tools dozens of times and reread the same files at each step, which is why agent work uses far more tokens than chat. Reasoning models also write thinking tokens that you may never see but still pay for.
What changes the count
- Language: many languages other than English take more tokens to say the same thing.
- Format: code, tables, JSON and unusual formatting tend to use more tokens per word than plain prose.
- Model: each model family has its own tokenizer, so the same text can come out to a different count on different models.
Keeping token use under control
Send only what the task needs, reuse unchanging context with prompt caching, and send routine work to cheaper models. Prices per token vary enormously between models, so which model handles a task usually matters far more than trimming words from a prompt.
How elvex relates
elvex Enterprise routes work to lower-cost models where they do the job, and admins can set budgets so token spend doesn’t surprise anyone at the end of the month.