Quick answer
Tokens are pieces of text a model processes. Credits are a billing unit. A quota is an allowance over a period. A window is that period. Cache reads are reused input billed at a lower rate, and reasoning tokens are internal output. A reset is when a window refills; it is not a refund.
The terms
| Term | Means | Where you see it |
|---|---|---|
| Token | Text unit, roughly a word fragment | Every usage report |
| Input token | Text sent to the model | Session logs |
| Output token | Text the model generates | Session logs |
| Cached input | Reused prefix at a lower rate | Cost breakdowns |
| Reasoning token | Internal work on some models | Usage panels |
| Credit | Billing unit | Balances and invoices |
| Quota | Allowance for a period | Plan pages |
| Window | The period itself | Reset timers |
| Reset | Refill of a window | Usage panel |
| Overage | Usage past the allowance | Billing |
The three confusions
- Tokens versus credits: one counts work, the other bills it. The exchange rate depends on the model.
- Context versus usage: one limits a request, the other limits a period.
- Reset versus refund: a reset restores capacity; it never returns money.
Use the terms precisely
When reporting usage, say which unit you mean and which source produced it. 'We used two million tokens' and 'we spent 400 credits' are both true and not comparable. UsageAtlas keeps the units separate for 14 providers.
Questions, answered
What is the difference between tokens and credits?
Tokens measure text processed; credits are a billing unit that converts model usage to a single number at model-dependent rates.
What is a usage window?
A period over which an allowance applies, such as five hours or a week. When it resets, capacity returns but nothing is refunded.
Are cached tokens cheaper?
Yes. Reused input is billed below the fresh input rate, and cache writes carry a one-off surcharge.