Skip to content
LLM Engineering

What is Token Metering?

Per-leg usage accounting that turns raw token counts into billable credits using each model price table plus markup. BYOB meters input, output, cached, and thinking tokens separately, with a minimum charge flooring tiny calls.

Example

When a session uses a premium tool model, the meter shows input, output, cached, and thinking legs priced from that model config, so swapping to a cheaper chat model visibly drops the next turn credit cost.

What people get wrong

Reading the credit total as exact provider spend. Credits add platform markup and minimum charges, so they track relative session cost, not the provider invoice.

Frequently asked questions

Why does one long session cost more than several short ones?

Long sessions accumulate output and thinking tokens and can cross long-context price thresholds, while short sessions reset the context each time.

Sources

Browse all LLM Engineering terms →