PRICING GUIDE

Compare AI API prices using the correct billing unit

Text, image, and video models measure usage differently. Normalize the workload before comparing rates or estimating production cost.

Updated 2026-09-11

In brief

TokenAAS normally prices text by input and output tokens, images per request or image tier, and videos per generated second with optional resolution tiers. Live configured prices appear in the model catalog.

Text model billing

Text usage normally separates input and output token rates. Cached reads or writes may have separate prices where supported.

  • Estimate prompt and completion tokens separately
  • Check context-dependent intervals where configured
  • Use streaming for responsiveness, not as a billing shortcut

Image model billing

Image routes use per-request prices or size-specific tiers. A model can reject dimensions that do not satisfy its pixel constraints.

  • Confirm supported size before production
  • Price multiple generated images as multiple requests unless documented otherwise
  • Store returned assets before temporary links expire

Video model billing

Video routes are asynchronous and normally use generated duration multiplied by the configured per-second price for the requested resolution.

  • Validate duration and resolution
  • Track failed-task refund behavior
  • Include polling and authenticated download in acceptance tests

Effective account price

An API key belongs to a model group. Group rate, account rules, promotions, and commercial terms can affect the effective amount shown in usage records.

  • Check balance before and after a controlled request
  • Compare usage records with the catalog unit
  • Use the live catalog as the current public reference

Frequently asked questions

Are catalog prices the same for every account?

The public catalog shows configured public rates. Account group and commercial settings may affect the effective price visible after sign-in.

Can text, image, and video prices be compared directly?

Not without normalizing the workload. They use different units: tokens, generated images, and generated video seconds.