On-demand, spot or a prepaid term for a fine-tune

On demand means you start a GPU when you want it and stop it when the run ends. You pay for the time it ran. For a single fine-tune, or a week of tries, this is the right way to rent. The hourly price is the highest of the three, and nothing is wasted.

Spot is spare capacity at a lower price. The machine can be taken back at short notice. A fine-tune can live with that if it saves checkpoints often and resumes from the last one without your help.

Count what an interruption costs: the work since the last checkpoint, the time to get a new machine, and the time to load the model again. If your run is interrupted often, the lower price can vanish.

A term is a commitment for a month or more, at a lower hourly figure, often paid ahead. You pay for every hour of the term. It suits a team that fine-tunes all the time. It does not suit one project with a few runs.

Multiply the term rate by every hour in the term. Compare that with the on-demand cost of the GPU-hours you would really use. The term wins only when you use most of it.

Estimate with on-demand prices first. They compare across providers without reading the terms. On 12 Nov 2026, on-demand H200 prices listed here ran from $3.49 to $10.60 per GPU-hour. Treat spot and term prices as a later saving, after the first runs have told you your real hours.

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.

This is one of a coordinated collection of Runcrate references. Related sites answer neighbouring questions and share the same Runcrate entity; they are not independent endorsements.