Light Logo

Billing

Understand how WriftAI billing, credits, and compute usage work.

WriftAI charges based on compute usage.

Pricing depends on the hardware used by the model you are running and is listed on the pricing page. Hardware information for each model is available on the model’s detail page and in the model object returned by the API.

What is billed

Billing applies to compute-backed operations, such as running a model.

Other operations (including browsing models, reading metadata, API requests that do not run compute, and local runs) are not billed.

How compute time is measured

WriftAI bills for the time a compute instance is actively running your workload.

Cold starts are not billed

If a model is not already running, WriftAI may need to spin up an instance. Time spent waiting for the instance to start is not charged.

Billable time

Billing only begins once the instance is ready and may include:

  • Loading model weights
  • Initializing the runtime
  • Any active processing after the instance is available

Billing granularity

Compute usage is billed per second.

  • All charges are rounded up to the nearest second
  • Partial seconds are billed as a full second

Credits

WriftAI uses a prepaid credit system. Credits are consumed as you run models.

Auto-reload

You can configure auto-reload to automatically add credits when your balance drops below a configured threshold.

Auto-reload settings include:

  • A minimum balance
  • A reload amount

Viewing usage and charges

You can purchase credits, view your credit balance, compute usage, billing history and invoices from the Billing page on your dashboard.

Credit validity

Purchased credits are valid for 1 year from the purchase date and are non-refundable.

Charges exceeding available credits

In rare cases, compute usage may exceed your available credit balance. When this happens, the excess amount is charged to the default payment method on file.