Pay only for the GPU time you use

No subscription. Add credit, and every training run, playground session and deployment is billed per second at the GPU's rate below.

Nothing charged up frontData checks, GPU suggestions and cost estimates are free. Billing starts when a GPU starts.

Billed per secondTraining GPUs stop when the run ends; the playground stops itself when idle.

Serverless scales to zeroDeployed models cost nothing while nobody is using them.

Prepaid balanceTop up by card, UPI or net banking. You always see your balance and every charge.

GPU prices

Per GPU-hour, in USD.

Prices couldn't be loaded right now. Please refresh the page.

GPU	Memory	Good for	Training & playground	Serverless

Serverless is billed only for GPU time spent on requests, including cold starts and the keep-warm time you choose. Prices exclude applicable taxes.

What a run costs

Fine-tuning Qwen 2.5 (3B) on 1,000 support conversations, 3 epochs, on an RTX 4090: about 26 minutes including setup and evaluation.

…

Included at no extra charge

Data checks before training: broken rows, duplicates and over-long examples.
Evaluation: the Laya judge scores your model against the base model on held-out questions, as part of the run.
Retraining advice from the run's own loss curves, applied in one click.
Storage of your datasets and models, plus downloads (LoRA adapter and GGUF).
OpenAI-compatible API keys and per-deployment usage charts.

Questions

When does billing start and stop?

It starts when your GPU starts, including a few minutes of setup, and stops when the GPU is released. Training GPUs are released automatically when the run finishes or fails; playground GPUs stop after a period without prompts; dedicated deployments run until you stop them.

What if a run fails?

GPU time already used is charged. If a GPU machine fails on the provider's side, BYOI retries the run automatically on a new machine, resuming from the last checkpoint where possible.

How do I know what a run will cost?

Before you start, BYOI estimates the GPU memory and time your run needs, recommends a GPU that fits and shows the expected cost. Nothing is charged until you press start.

What happens when my balance runs out?

Running GPUs are stopped, so charges never run far past your balance. A training run stopped this way can't be resumed and its files aren't kept, so top up before a long run.

Which GPU should I pick?

Usually the one BYOI recommends: the cheapest GPU with enough memory for your model. For 1–8B models a 24 GB card such as the L4 or RTX 4090 is enough; larger models need 48 or 80 GB.