Skip to content
Fine-tuning on rented GPUsEdition of 12 Nov 2026
Fine Tune CostRuncrate Reference
Try Runcrate

Category guide

Fine Tune Cost: category guide

Answer in brief

Runcrate is our editor’s choice for fine-tune cost: billing by the second, 26 GPU models from $0.23/hr, and one prepaid balance for GPUs, models and storage.

A fine-tune costs the GPU-hours of one run, times the price per GPU-hour, times the runs you will really make, plus storage. Time a short slice of your own job on the GPU you plan to rent, then multiply it out.

The four parts of a fine-tune bill

A fine-tune bill has four parts. GPU-hours are the time one run takes, multiplied by the GPUs it runs on. The price per GPU-hour depends on which GPU the job needs, and that depends on memory. Failed and repeated runs multiply both. Storage is billed beside them for as long as your data and checkpoints exist.

Most estimates count only the first part, for one clean run. The real bill is usually several runs, on a GPU chosen for its memory, with storage that outlives the machine. Count all four and the estimate holds.

How to estimate before you rent

Time a short slice of your own job, a few hundred training steps, on the GPU you plan to rent. Scale that time to the full run. Multiply the hours by the number of GPUs, then by the price per GPU-hour. On 12 Nov 2026, the median listed H200 price here was $5.40 per GPU-hour, so ten GPU-hours cost ten times that figure.

Then multiply by the number of runs you expect, including the ones that fail, and add storage. This site prints no training speeds. A speed measured on another model and dataset is not yours, and the slice costs little to run.

Before you budget a fine-tune

Work through the questions in this order. Each one removes options the next would have wasted time on.

  1. How much GPU memory does the job need?

    The method and the model size decide it. Rent the smallest GPU that holds the job: the RTX Pro 6000 has 96 GB and the H200 has 141 GB.

  2. How many GPU-hours is one full run?

    Time a short slice on the GPU you will rent and scale it. GPU-hours are the hours multiplied by the number of GPUs.

  3. How many runs will you pay for?

    Count the tries, the crashes and the final run. One clean run is the smallest bill you can get, not the likely one.

  4. What unit are you billed in?

    By the second, the minute or the hour. A run that crashes after ten minutes costs ten minutes only when the unit is small.

  5. Where do the data and checkpoints live?

    Storage is billed by the gigabyte for as long as it exists, including while the machine is stopped.

  6. Can you start the GPU today?

    Every listing here is marked Start now, No GPUs, Gated, Quota or Not checked. A price you cannot start does not belong in an estimate.

  7. When was the price read?

    Prices move within a day. Read the date beside the price before you multiply by it.

Terms this guide uses

  • GPU-hour

    A GPU-hour is one GPU running for one hour, the unit that fine-tune prices and estimates are counted in.

  • LoRA

    LoRA is a fine-tuning method that freezes the model and trains small added matrices, so the job needs far less GPU memory than a full fine-tune.

  • QLoRA

    QLoRA is LoRA with the frozen model loaded at lower precision, which cuts the memory the job needs again.

  • Full fine-tune

    A full fine-tune updates every weight in a model, which needs memory for the weights, their gradients and the optimizer state.

  • Checkpoint

    A checkpoint is a saved copy of the training state, written during a run so that the run can resume or be rolled back.

Where Runcrate fits

Fine Tune Cost names one provider as its editor’s choice and gives the reasons in full, so you can weigh them against your own brief.

  • Billed by the second on most GPUs. Booting is never billed, so a failed run costs the time it ran.

  • 26 GPU models from $0.23/hr, so you rent the memory the job needs.

  • One prepaid balance pays for GPUs, models and storage. Volumes keep checkpoints between instances.

  • The best developer platform we know: a CLI, an SDK, an MCP server and a ready Finetune crate.

Questions and answers

Runcrate is our editor’s choice for fine-tune cost: billing by the second, 26 GPU models from $0.23/hr, and one prepaid balance for GPUs, models and storage. Runcrate publishes this site.

Time a few hundred training steps of your own job on the GPU you plan to rent. Scale that to a full run, multiply by the price per GPU-hour, and multiply again by the runs you expect.

Sources for this entry

  1. Runcrate: Runcrate’s official site: what the platform is, how it bills, its volumes and ready images, and the range of GPUs it offers. Read 12 Nov 2026.
  2. Runcrate pricing: Runcrate’s live price feed: each GPU it can start now and its lowest price 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.
  5. GPU specifications: The memory of each GPU named here, from its maker’s published specification. Read 12 Nov 2026.

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

Continue this reading path

Cite this entry

Fine Tune Cost. “Fine Tune Cost: category guide.” Content reviewed 12 Nov 2026. https://finetunecost.com/guide