Docs
Billing and spend limits
How does Onrup billing work?
Compute is metered per GPU-second above a one-minute floor. Before any job starts, its cost is estimated and reserved against your spend limit — counting open reservations as well as settled spend — so the limit is a gate rather than an alarm.
Authorised, then spent
The sequence is: estimate the maximum cost, reserve it against available headroom, lease compute. If the reservation cannot be made, no compute is leased and the call returns a specific error.
The alternative — measuring spend and alerting when it exceeds a threshold — tells you about an overrun after the money is gone. That is a notification, not a limit.
Headroom counts reservations
Available budget is your limit minus settled spend minus open reservations. Without counting open reservations, several jobs authorised at the same moment could each pass the check individually and collectively exceed the ceiling.
Reservations expire after a fixed window, so an abandoned job does not hold budget indefinitely.
{
"plan_id": "pro",
"subscription_status": "active",
"allows_compute": true,
"spend_limit_cents": 50000,
"reserved_cents": 1840,
"settled_cents": 12655,
"available_cents": 35505
}Metering
- Training compute — per GPU-second, one-minute minimum. Time waiting for capacity is not billed, and neither is teardown.
- Serving compute — per GPU-second while a replica is live, whether or not requests arrive.
- Storage — per GB-month above the plan allowance, covering datasets, checkpoints, weights and evaluation artefacts.
- Evaluation and generation — per job, against a ceiling declared before the job starts.
Delivery lag is shown, not hidden
Usage events take time to reach the billing system. The usage endpoint reports pending events separately from settled ones rather than showing a single figure that pretends the gap does not exist.
Every event carries the pricing revision in force when it was recorded, so a historical invoice can be reconstructed exactly even after prices change.
Frequently asked questions
What happens when I hit my limit?
New work is refused with a specific error code. Running work continues to its natural end, because killing a run mid-flight would waste everything spent on it so far.
Do I pay while waiting for capacity?
No. Metering starts when your job is on the GPU. Queue time is not billed.
Start with the free tier
A magic link creates your account, your tenant and your first API key. No card until you ask for compute.