Quick answer
Premium requests counted calls; AI credits price them. Under the credit model, a request to an expensive model with a large context costs more than the same request to a cheap model. That makes your remaining balance a better guide than a request count, but it also makes cost less predictable.
What changed
| Dimension | Premium requests | AI credits |
|---|---|---|
| Unit | Request | Currency |
| Cost per call | Multiplier | Tokens x rate |
| Long context | Often flat | More expensive |
| Agent loops | One per request | One per model call |
| Planning | Count requests | Watch the balance |
What stayed the same
- Default-model completions remain cheap or included.
- The allowance resets on the billing period.
- Organization admins still control enablement and budgets.
- Usage is still reported per seat and per model.
Read your account correctly
Look at the unit label before comparing screenshots or advice from a year ago. If your account shows requests, plan in requests; if it shows credits, plan in credits. Converting between them with one number will mislead you.
UsageAtlas displays whichever unit the provider reports, so the dashboard does not mix the two.
Why both names appear
Premium requests describe consumption: a model call with a premium model. AI credits describe the billing unit that some plans use to pay for that consumption. One is a count, the other is a currency, and the exchange rate depends on the model.
| Question | Look at |
|---|---|
| How many calls did I make? | Premium requests |
| What did they cost? | Credits or billing |
| How many are left? | The unit your plan reports |
| Can I buy more? | Billing and overage settings |
Track the unit that blocks you. The credits guide covers balances, and remaining credits covers where to read them.
Questions, answered
Did Copilot stop using premium requests?
Newer plans use AI credits. Some accounts still show premium requests during the transition, so check your own settings.
Is one premium request equal to one credit?
No. Credits vary with model and tokens, so there is no single conversion rate.
Which model is cheaper under credits?
The default or smallest model, especially for short tasks. Premium models cost more per token and per call.