{"verified":{"vast":{"billing_granularity":"per-second while the instance is active","storage_when_stopped":"BILLED — 'stopping an instance does not avoid storage costs'; rates are host-set","bandwidth":"BILLED both directions, per byte, at host-set rates — NOT included in the listed $/hr","gotchas":["upload/download spend does not appear in the $/hr or $/day views — the classic 'credit drained faster than expected' surprise"],"source":"https://docs.vast.ai/billing","verified":"2026-08-24"},"runpod":{"billing_granularity":"per-second for pods and container/volume disk; network volumes bill hourly","storage_when_stopped":"volume disk $0.20/GB/mo while STOPPED — double the running rate ($0.10); network volumes $0.05–0.07/GB/mo either way; container disk free when stopped","bandwidth":"none — 'no fees for data ingress or egress'","gotchas":["balance hitting $0 terminates pods without network volumes and the data is unrecoverable","savings plans cover GPU compute only — storage bills at standard rates regardless"],"source":"https://docs.runpod.io/pods/pricing","verified":"2026-08-24"},"hyperstack":{"billing_granularity":"per-MINUTE — coarser than the per-second rails","storage_when_stopped":"storage billed continuously (~$0.10/TB/hr); the pricing page does not publish a stopped-VM policy","bandwidth":"free — 'Egress/Ingress traffic: Free'","gotchas":["public IPs are metered separately (~$0.0067/hr)","snapshots are billed; the cost is only estimated at creation"],"source":"https://www.hyperstack.cloud/gpu-pricing","verified":"2026-08-24"},"datacrunch":{"billing_granularity":"NOT DISCLOSED on the public pricing page (which now serves at verda.com)","storage_when_stopped":"storage priced per GiB-month; stopped-instance policy not published","bandwidth":"not disclosed","gotchas":["undisclosed fine print is itself the warning — ask before committing a long job"],"source":"https://verda.com/pricing","verified":"2026-08-24"}},"unverified":["cudo","akash","ionet","novita","primeintellect","shadeform"],"note":"billing fine print read from each provider's own published pages on the date shown — quotes, not paraphrase, where it matters. Providers under 'unverified' get NO claims from us until we have read their pages. Fine print changes; the source link outranks this cache."}