Interruptible terms
Interruptible capacity costs less than Dedicated capacity because it can be taken back. These terms apply to every Interruptible launch, Task and reservation.
Version 2026-11-04
Notice
Interruptible: may be stopped at any time without notice. You are only billed for what you actually use.
What Interruptible means
Interruptible capacity can be stopped at any time, without notice, when the capacity is needed for a Dedicated order or a reserved slot. A stop signal may be sent to your workload before it stops, but it is never guaranteed.
What you pay
You are only billed for what you actually use: billing stops at the exact millisecond the capacity is taken back. The hourly price is fixed when you launch and never changes while it runs.
By Rental Offer
- On demand: the instance is stopped. You receive an email and a webhook event.
- Time slicing: the share is stopped, or paused without being deleted while Dedicated shares on the same card need the compute. Paused time is not billed.
- Pay per task: the Task goes back to the queue and runs again, up to the number of retries you set. A new run starts your command from the beginning, so save checkpoints. Compute time already used is billed once; waiting and restarts are not billed.
- Reserved: a slot taken back before it starts is refunded in full. A slot taken back while it runs is billed for the time served only; the rest is refunded.
Your data
Data stored on a stopped Interruptible instance may be lost. Keep checkpoints and outputs outside the instance, in a volume or at an address you control.
Acceptance
Before your organization's first Interruptible launch, a member accepts these terms. The acceptance is recorded with its date and the version of the text. A new version is accepted again before the next Interruptible launch.
