# Credits

## Overview

Hub AI credits allow for the compute used by training and evaluation runs. Hub shows a conservative estimate before submission,
reserves that amount for the run, and settles the balance from measured usage when work ends.

> **Note**
> Credit balances belong to the active Hub team. Check the team balance before starting a run, particularly when other teammates may be running experiments at the same time.

## Check the team balance

Open Team Settings and select Subscription. The AI credits card shows Available credits that can be reserved for new work and
Reserved credits held for active or queued runs.

The same card groups the balance by source and shows expiry dates where applicable. Use those details to understand which credits
are temporary and when the team's available balance will change.

See [Team & Members](https://docs.luxonis.com/cloud/organization/team.md) for the other subscription, billing, and quota settings.

## How a run uses credits

The credit lifecycle has four stages:

 * Estimate: The training or evaluation form calculates a conservative estimate from the requested work and selected compute
   speed.
 * Reserve: When the run is submitted, Hub reserves the estimate. Reserved credits cannot be used by another run.
 * Measure: While the run is active, Hub measures compute usage.
 * Settle: When the run completes, Hub charges only the credits actually used and releases the unused part of the reservation.

The run page shows the settled Credits Used value.

> **The estimate is a reservation, not the final charge**
> The estimate intentionally leaves room for uncertainty. A normal completed run is charged for actual usage, not automatically for the entire estimate.

### Interrupted, failed, and stopped runs

| Outcome | Credit behavior |
| --- | --- |
| Completed | Actual compute usage is charged; unused reserved credits are released |
| Reservation exhausted | Hub interrupts the run when its reserved amount is exhausted; the best checkpoint produced so far is
retained |
| Failed | Reserved credits are refunded |
| Stopped by a user | Compute used before the stop is charged; the unused reservation is released |

Stopping early can reclaim unused reserved credits, but work already performed is still charged.

## Resolve estimate and balance problems

If Hub reports insufficient credits:

 * review Available and Reserved balances in Team Settings;
 * wait for active work to finish or stop a run that is no longer needed;
 * reduce the requested work when the form supports doing so;
 * reach out to support if the problem persists.

Training speed affects both duration and credit consumption. Review the compute type and estimate shown for each speed available
to your team before choosing one.