Model Pricing

AISIX Cloud reports the spend shown in usage and budgets by computing a cost for each usage event from its token counts and a per-model price. This page explains where those prices come from and how to set a price for a model the catalog does not already cover.

Use the Model pricing page when usage shows token counts for a model but its spend stays at zero, or when you need a custom rate for a self-hosted or reseller upstream.

Where Prices Come From

Every price is data — AISIX applies no hidden multipliers. A price is resolved from one of two sources, keyed on the pair (provider, model name):

• Catalog defaults come from models.dev and refresh automatically. They cover the well-known providers and models, so most catalog models are priced without any action.
• Organization overrides are prices you set yourself. An override wins over the catalog default for the same (provider, model name) within your organization.

Cost is computed when a usage event is recorded and stored with the event. Changing a price affects events recorded after the change; it does not re-price historical events.

Models Without a Catalog Price

The catalog only prices the (provider, model name) pairs models.dev publishes. A model has no catalog price when, for example:

• it runs on a self-hosted or reseller upstream that models.dev does not price, or
• its model name is a custom alias, or a newer model the catalog does not carry yet under that provider.

When no catalog default and no override match a model, AISIX cannot derive a cost, so that model's spend is reported as $0.00 even though its token counts are correct. To start reporting spend for such a model, set a price for it.

Set a Price

1. Open Model pricing from the organization navigation.
2. Select Add pricing.
3. Enter the Provider and Model name exactly as configured on the model — cost is matched to usage events by this (provider, model name) pair, so a mismatch produces a price that never applies. The provider field suggests known catalog providers, but you can type any value, including a custom one.
4. Enter the Prompt and Completion rates in USD per 1M tokens. Optionally set distinct Cache read, Cache write, and Reasoning rates.
5. Save. The price applies to that model's usage events from the next event onward.

Prices are USD per 1M tokens. Leave a cache or reasoning rate at 0 to fall back to the prompt or completion rate for those token classes.

Edit or Reset a Price

Each row on the Model pricing page can be edited. A row sourced from an override can also be reset to its catalog default, which removes the override so the catalog price takes over again on the next usage event.

Next Steps

See Usage Reporting for how the resolved cost appears in spend, budgets, and top-model views, then continue with Logging and Auditing to investigate individual request outcomes.