Estimate what it costs to train an ML model.

Cached prices · updated just now

What can this budget buy?

With a fixed budget, a bigger model means less data. Set your budget to see the trade-off and the best balance.

10,000

Covers GPU compute, tuning runs and storage.

Model size vs. data

Every point on the solid line uses your full budget. The star marks the best balance.

1.00B10.00B100.00B1.00T10.00T100.00M1.00B10.00B100.00BBest balance: 88.67B tokens, 4.43BYour selection: 89.13B tokens, 4.41BDataset size (tokens)Max model size (params)
Affordable with your budget Ideal balance (20 tokens per parameter)
89.13B tokens

More data, smaller model for the same spend.

Your selection

Max model size (params)
4.41B
Dataset
89.13B tokens
Tokens per parameter
20.2
Training time
6.27 days
Compute
$9,946
Storage
$54
Total
$10,000
Budget used
100%

of $10,000

Good balance. 20.2 tokens per parameter, in the optimal 10–30 range.

Run settings

Filled in from your setup. Your changes stay until you change the setup above.

Each tuning run as a share of a full run.

Recommended: 1, to finish within 60 days.

p5.48xlarge (8x H100 80GB) · bf16 · 55% GPU utilisation · $55.04/hr per instance · 8 GPUs total