Cost Settings
Costs are calculated and reported in USD, priced per model from the prices in the model's LLM provider.
Per-Model Pricing
Every model of an LLM provider can have an input and output price, expressed per 1M tokens:
gpt-4o
Input: $2.50 per 1M tokens
Output: $10.00 per 1M tokens
gpt-4-turbo
Input: $10.00 per 1M tokens
Output: $30.00 per 1M tokens
Pricing changes apply going forward — historical usage keeps the price that was in effect when it was recorded.
In the model drawer, the prices appear as Input $/1M, Output $/1M and Cached $/1M. Prices are optional. A model with no price shows a no price badge in Cost management of the model drawer.
Go to AI Gateway → Models and open the provider that serves the model. Pricing lives in the model's drawer, under Cost management.
The Models card lists the models of the provider. Click Configure on the card, then choose Edit in the "⋯" menu of the model's row to open the model drawer.
Edit the input and output price fields. A model with no price configured counts as 0 in cost reports.
Click Save Model, then Save in the Configure drawer and Confirm the dialog. New requests use the updated price; already-recorded usage is not recalculated.
Adding a Model
If the model you want to price is not on the provider yet (for example a fine-tuned model), add it:
In the Configure drawer of the provider's Models card, click Add. The Discover from provider tab fetches the provider's model list; the Add manually tab lets you type the model ID.
Enter the model ID (for example my-fine-tuned-gpt4o).
In the model drawer, set the unit prices for input, output and (if the provider offers it) cached input tokens.
Click Save Model, then Save and Confirm. The model is available for routing and reports.
Audio (speech-to-text, text-to-speech) and image generation models are priced in other units, such as per minute, per character or per image. See Multi-Modal Endpoints.
Cost Reconciliation
Every request's cost is reconciled from three token counts, priced independently:
Tokens sent to the model as the prompt.
Tokens generated by the model in its response.
Prompt tokens the provider served from its own prompt/context cache instead of reprocessing them, usually billed at a lower rate than standard input tokens.
Cached tokens are a provider-side mechanism (OpenAI prompt caching, Anthropic prompt caching, etc.) — not Apinizer's own semantic cache, which avoids calling the model at all for a similar prior request. A request can use provider-side caching and still be a full, billable call to the model.
Because each is priced separately, your cost reports reflect exactly what was consumed rather than a single blended rate.
Prices and USD Limits
The same prices also drive USD limits in Limit Plans, so the amount charged against a USD limit and the cost shown in reports are the same figure:
- Each turn of a request is priced by the model that actually served it. When a request fails over to another model, or is routed to a cheaper model, the cost follows the model that answered.
- A request served by a model with no price costs 0 in both places. The model is exempt from USD limits, and its usage appears in reports at zero cost, until you set a price. Check the no price badge in the model drawer.
- A request whose price resolves to 0 (for example, a zero-rate service tier) also counts 0 toward the USD limit.
Cost in Reports
Once pricing is configured, cost figures appear throughout Reports and Analytics:
- The Total cost card on Analytic → Dashboard (AI view)
- This month's cost and the cumulative cost chart on the Cost & Consumers tab of Reports → Metrics (AI view)
- Cost breakdown by person, group, model, project, team, and deployment type on the Usage Reports tab of the same screen
- Exported reports with cost columns
- The Cost column on Analytic → Traffic (AI view)
Reports show costs in USD, with four decimals.
Permissions
Prices are edited together with the LLM provider, so changing them requires the permission to manage AI development. A provider shared from the Administration catalog is edited from the Administration menu. Other roles continue to see usage and cost breakdowns in reports, without the ability to change pricing.
Troubleshooting
Cost Looks Too Low or Zero in Reports
- Check that the model has a price on its provider
- Models with no price count as 0; set the price in the model drawer
Old Cost Data Looks Wrong
- Price changes are not applied retroactively; only new requests use the new price
- For past costs, look at the report of the relevant period