Runpod serverless pricing and other GPU clouds

Compare published per-second GPU rates with hourly instance prices. Idle workers, cold starts and minimum billing rules can change the bill; the label “serverless” does not establish those rules. Rates below show both units; the published unit and rounding differ by provider.

Runpod Flex workers are billed from startup until fully stopped, rounded up to the nearest second. That includes loading the model, request execution and the idle timeout. Active workers stay running and require a sales inquiry for discounted rates. The table includes only individually named GPUs; mixed-card classes cannot promise a particular card.

Runpod publishes hourly equivalents on its pricing page; their per-second values here are those figures divided by 3,600 and may be rounded. Modal's per-second GPU rates are multiplied by 3,600 for comparison. Separately billed CPU, RAM and storage are excluded.

When serverless wins

Modal includes a recurring monthly free credit tier; see free credits. RunPod also offers serverless endpoints priced separately from the pod rates shown on the RunPod page.

When does an always-on GPU cost less?

For the same GPU variant, divide the single-GPU instance rate by the serverless hourly equivalent. That fraction of an hour is the compute-only crossover. Request overhead and separately billed CPU, RAM, storage or warm workers can move it.

At a request duration you supply, requests at crossover = active seconds ÷ billed seconds per request. This assumes no overlap between requests and no idle worker charges. Per-second GPU pricing does not by itself guarantee zero idle charges.