How Tokens Work

Models process text by splitting it into smaller units called tokens. A token can be:

• part of a word (for example, "techno" + "logy");
• a short word;
• a symbol or punctuation mark.

On average, 1,000 tokens ≈ 750 words.

These values are approximate and provided only to give a general idea of usage costs. Each model uses its own tokenizer and splits text differently.

Tokens are billed for both input and output. For example, if your request contains 20 tokens and the model generates a response of 30 tokens, a total of 50 tokens will be deducted from your quota.