Quick answer
Tools measure usage in one of four units: tokens processed, requests sent, credits spent, or time on task. Tokens describe model work, requests describe calls, credits describe billing, and time describes agent runtime. None converts cleanly into another, so compare within a unit or measure cost per completed task.
The four units
| Unit | Measures | Use it for |
|---|---|---|
| Tokens | Text processed | Cost and context |
| Requests | Model calls | Plan allowances |
| Credits | Billing units | Budget |
| Time | Agent runtime | Orb-style metering |
Why conversion is lossy
One request can carry thousands or millions of tokens, depending on context. One credit can buy a different number of tokens depending on the model. That is why a chart that mixes units is unreadable even when each number is correct.
- Pick the unit of the constraint that blocks you.
- Track that unit consistently.
- Convert to cost only when you have a rate.
- Label every estimate as an estimate.
Compare across tools
The only cross-tool comparison that survives unit differences is cost per completed task. UsageAtlas normalises token fields where tools record them and shows reported plan usage where they do not, so both halves are visible. The token types guide defines each token type.
Questions, answered
Are tokens and requests the same?
No. A request can carry any number of tokens. Plans often limit one and bill on the other.
What is the best metric for comparing tools?
Cost per completed task. Raw tokens and request counts both favour different workflows and neither maps to value.
Why do providers use credits instead of tokens?
Credits let a provider price different models and features on one scale without exposing per-token economics.