How it works
Tokenisation breaks text into sub-word units using a vocabulary the model was trained on. Common tokens (the, of, ing) are single tokens; rare or specialised words may take multiple tokens. Pricing scales with token count, so efficient prompting matters at scale.
Example
A 1,000-word page is roughly 1,300-1,500 tokens. At Claude Opus 4.7's $5/M input rate, that's $0.0065-0.0075 per page processed.