Compute

One interface for cloud GPUs.

Point Compute at a Python function. It provisions a fresh machine for the job and streams output to your terminal. When the run ends, Compute confirms termination and puts provider usage and the platform fee on one receipt.

Install the CLISee pricing
Fine-tuning modelsReinforcement learningBatch inference

A run, end to end

simple_mlp.py · H100live

$ compute run simple_mlp.py::train --gpu H100

  1. estimaterate locked · confirmation required
  2. provisionfresh H100-SXM requested
  3. ready80 GB HBM3 available
  4. stdoutstep 1/3 · loss 0.8421
  5. result{"status":"ok","artifact":"result.json"}
  6. terminatedmachine gone · receipt final
Provider usage$0.80
Platform fee$0.06
Total debit$0.86
Providers

Choose where the GPU runs.

RunPod, Hot Aisle, and Vast.ai are available now. AWS, GCP, and Azure are coming soon. See the GPU list.

  • RunPodSecure H100 capacity
    Available
  • Hot AisleMI300X capacity
    Available
  • AWSEC2 GPU instances
    Coming soon
  • GCPCompute Engine GPUs
    Coming soon
  • AzureAzure GPU VMs
    Coming soon
  • Vast.aiSpot instances
    Available
Get started

Start a run in four steps.

  1. 01

    Sign up

    Use an enabled sign-in method.

  2. 02

    Buy credits

    Add at least $10 in prepaid credit.

  3. 03

    Install the CLI

    Install directly from compute.cx.

    curl -fsSL https://compute.cx/install.sh | sh
  4. 04

    Run your computation

    Pass a Python entry point and choose a GPU.

    compute run simple_mlp.py::train --gpu H100
Use cases

Find the guide that fits.

Each guide explains the workload and gives you the exact command.

SFT

Fine-tune a model

What
Train an open model on examples from your task.
When
The base model works, but it misses your tone, output format, or domain.
Run
compute run mi300x_lora_finetune.py::finetune --gpu MI300X
A 70B fits on one 192 GB card.
Open guide
RL

Reinforcement learning

What
Use a reward signal to improve outputs when labeled examples are not enough.
When
You can score an answer more easily than you can write the ideal one. Typical jobs rank outputs or train reasoning agents.
Run
compute run rl.py::train --gpu MI300X
Use --wait, or detach and follow logs later.
Open guide
Batch

Batch inference

What
Run the same model across a large input set and collect the outputs.
When
Useful for evals, embeddings, synthetic data, and overnight labeling.
Run
compute run batch_infer.py::generate --gpu MI300X
Pay only for the minutes the machine exists.
Open guide
Receipt-first pricing

Provider rate, plus 7.5%.

Each run uses the provider rate locked when you request the machine, plus a flat 7.5% platform fee. There are no subscriptions or usage tiers. A machine that never becomes ready costs $0.

Full pricing
One run, one machine

Run the function and keep the result.

Create an account and add credit. Once the CLI is installed, the first GPU run is one command away.

Install the CLISign in