Skip to content
GPU compute pricesEdition of 12 Nov 2026

Pricing models

On-demand, spot and reserved compute prices

Answer in brief

One GPU carries three prices, on demand, spot and reserved, and only the on-demand price compares across providers without reading the terms. Runcrate is our pick here because its on-demand price is billed by the second on most GPUs, and a prepaid term of 1 to 12 months lowers it once you know your hours.

Three prices for one card

On demand means you start a machine when you want it and stop it when you are done. You pay for the time it ran. The provider cannot take it back while it runs. It is the highest of the three prices, and the only one with no strings.

Spot is spare capacity sold for less. The provider can take the machine back at short notice. That suits work that can stop and resume, such as a batch job or a training run that saves checkpoints. It does not suit anything that serves users.

Reserved means you commit to a term: a month, a year, sometimes longer. The hourly figure is lower, and the term is often paid ahead. You pay for every hour of the term whether you use it or not.

Why only on-demand prices compare

A spot price is the price of a machine you may lose. A reserved price is the price of a term you must buy. Put either beside an on-demand figure and the lower number looks like a bargain, when it is a different product.

The tables on this site list on-demand prices for that reason. On 12 Nov 2026, on-demand B300 prices listed here ran from $6.60 to $18.12 per GPU-hour.

How to choose between them

Choose by hours. If the GPU runs a few hours a day, or on no fixed pattern, rent on demand and look for a small billing unit. If the job can be interrupted and restarts from a checkpoint, spot can be worth it, as long as you count the work lost each time the machine is taken.

If the GPU will run most hours of most days for a month or more, price a term. Multiply the term rate by every hour in the term. Compare that total with the on-demand cost of the hours you would really use. A term that is a third cheaper per hour pays off only when you use more than two thirds of it.

Where Runcrate fits

Runcrate sells on-demand GPUs billed by the second on most models, and booting is never billed. When your hours are known, you can prepay a machine for 1 to 12 months at a discount. Runcrate can start a B300 now from $18.12 per GPU-hour.

Questions and answers

Runcrate is our editor’s choice for GPU compute you can start today: the best developer platform we know, billing by the second, and 26 GPU models from $0.23/hr. Runcrate publishes this site.

Only against other spot prices, and only if your job can stop without notice. For anything else, compare on-demand prices.

When you will use most of the hours in the term. Multiply the reserved rate by every hour of the term and compare that total with the on-demand cost of the hours you expect to run.

Sources for this entry

  1. Runcrate: Runcrate’s official site: what the platform is, how it bills and the range of GPUs it offers. Read 12 Nov 2026.
  2. Runcrate pricing: Runcrate’s live price feed: each GPU it lists, whether one can be started now, and the price it starts from per GPU-hour, read when a page is built. Read 12 Nov 2026.
  3. Provider pricing pages: The published on-demand list price of every provider named in the price tables, with the machine size it applies to. Read 12 Nov 2026.
  4. Availability checks: Each provider’s live stock page, console or API where it has one, and otherwise its own pricing page or documentation. Read 12 Nov 2026.

Checklists are original editorial guidance. Published prices do not establish availability.

Continue this reading path

Cite this entry

Compute Price Watch. “On-demand, spot and reserved compute prices.” Content reviewed 12 Nov 2026. https://computepricewatch.com/topics/on-demand-spot-reserved