What you pay for

You pay for every second a provider bills for a machine Nodus acquired for your work, from the moment billing starts until the machine is confirmed deleted. Nodus pays for everything it chose or caused. Your usage shows each machine’s time by segment, so you can see where every second went:

• Boot: from billing start to your command starting.
• Restore: a replacement machine’s time until your command starts again after a recovery, including the checkpoint restore.
• Running: from your command starting to its stop.
• Teardown: from stop to confirmed deletion, plus the provider’s billing increment, rounded once per machine.

Run nodus get usage --group-by segment or open Usage & billing → Usage in the console to see the split. For a multi-node run, --group-by rank,segment itemizes each member.

The table

What	Segment	Who pays
Time from billing start to your command starting: boot, registration, readiness checks, image pull and checkpoint restore	Boot, Restore	You
A boot that fails on the provider’s side, until the machine is confirmed deleted	Boot	You
Spare machines Nodus starts to finish your work sooner, machines Nodus releases or replaces on its own decision, machines that fail Nodus’s checks, and failures Nodus causes	None: never on your usage	Nodus
From your command starting to its stop, including lazy image fetches, checkpoint writes, the snapshot on stop and work lost after a preemption	Running	You
Downloading inputs after your command starts, and the initCommand preflight	Running	You
Imports, exports and sink loads that run on your org’s capacity	Running	You
Disk above 10 GiB per vCPU on Nodus nodes	Running	You
A network partition your work tolerates, until it reconnects or until the old machine is fenced and confirmed deleted	Running, Teardown	You
The provider’s billing increment, rounded once per machine	Teardown	You
Warm idle time beyond an increment you already paid for, minimum workers, and idle Function and agent workers	Warm idle	You
The idle share of Nodus-operated capacity, built into its rates and never metered on its own	Included in rates	You
From stop to confirmed deletion	Teardown	You
Time between stop and the delete request beyond 60 seconds when Nodus caused the delay	None: never on your usage	Nodus
Any cost beyond your tightest limit: maxCostUSD, a Budget or your balance	None: never on your usage	Nodus
A model request whose outcome Nodus cannot confirm, and upstream charges beyond the usage observed on a cut-off stream	None: never on your usage	Nodus
Console assistant and command generation calls, within the daily cap	None	Nodus
Grading Sandboxes, evaluations, dataset previews, task manifests and model calls made from inside your containers, under your caps	Running	You
Mirroring a private image so it can run on hosted containers, and the mirror’s storage	None	Nodus
Nodus’s own test runs	None	Nodus
Image builds that fail	Build	You
Storage above 10 GB per org: Volumes, checkpoints, outputs, images and session state	Storage	You
Logs	None	Nodus
Egress within 10 GiB per org per day	None	Nodus
Egress beyond the included 10 GiB a day, up to your quota, and relayed traffic between multi-node members	Egress	You
Your own pool hosts and cloud accounts	None	Free
Pool devices running work Nodus schedules, Predict per pool, and capacity Nodus acquires for you	Platform	You
Differences between a provider’s invoice and what Nodus metered	None	Nodus

This table is checked against the billing reference model on every change, so the rows here are the rows Nodus charges by.

Limits stop work before it overruns

Every paid action holds funds before it starts, and work stops gracefully when a hold cannot be renewed. If a charge would still go past your tightest limit, Nodus absorbs the difference: your balance never pays more than the limit you set. See How billing works for holds, captures and the low-balance stop.