Tokens (AI usage)
The chunks of text, and images converted into similar units, that AI providers count and bill for, in both what you send and what comes back.
Updated October 2026 · with examples from our own builds
What does tokens mean in practice?
AI providers charge per token, roughly a fragment of a word, for both input and output, rather than per request. Images count too, often as many tokens. Long prompts, long histories and large photos all raise the cost of a single call.
Tokens are the unit of AI cost, and two features that each make one call can differ in cost many times over. The only reliable figure is a measured one: record the tokens each call reports, price them, and add them up per user or per plan.
They also cap what fits in one call. A model can only consider so much at once, so products working with large amounts of text have to narrow it down first.
Tokens at a glance
- Billed on
- Input and output tokens, priced per model
- Measure
- Tokens reported by every call, per user per day
- Also limits
- How much text fits in a single call
Where has tokens come up in our builds?
The token usage from every AI response is recorded per user per day, priced in micro-dollars against a per-model table and summed into an AI cost card for the last 30 days. Background work with no user to bill goes to a separate system bucket.
On a 15-form benchmark the pipeline built 14 of 15 forms cleanly on a cheaper model that used about 30% fewer tokens, against 15 of 15 on the main model.
Discovery read 1,475 sentences off Storafleet's pages, far too many to show a model at once, so code narrows each flagged sentence to a handful of candidate claims before the single model call.
Common questions about tokens
Why does one AI feature cost so much more than another?
Their token counts differ. Photos, long conversation histories and long instructions all add input tokens, and some models count images as far more tokens than others, so measure each feature's tokens rather than comparing prices per request.
Should users see token counts?
Usually not. Show them counts they understand, such as scans or messages a day, and keep tokens and cost as internal measures. LeafRanger users see scans and messages; the admin dashboard sees the cost.
Related terms
Planning a product and not sure where tokens fits? Ask us, or browse all 116 terms in the glossary.