Models and pricing
Inference usage is billed per 1 million tokens. Rates vary by model and token type:
- Input: Tokens included in your prompt
- Cached input: Input tokens served from cache
- Output: Tokens generated by the model
The same rates are shown per model under Inference API → Models in the Aurora Portal.
Available models and pricing
The table is loaded from the live model catalog of the selected inference endpoint, so it always reflects current availability and rates. Prices are in USD per 1 million tokens.
| Model ID | Name | Context length | Input $/1M | Cached input $/1M | Output $/1M | Modality | License |
|---|---|---|---|---|---|---|---|
Loading model catalog from GET https://ai.aur.lu/v1/models… | |||||||
To read the current model list and prices programmatically, call GET https://ai.aur.lu/v1/models (no authentication required).
The full response format is in the API reference.
Aurora hosts these models as-is; each model remains subject to its upstream license, linked in the table. The GET /v1/models
response includes each model's license, license_url, model_card and pricing (USD per token).
Use the exact model ID returned by the catalog in your requests, including its author prefix.
You can also use any compatibility alias in the model's aliases array.
Compatibility aliases
Compatibility aliases keep older model IDs callable after a model update. They resolve to the current model and share its availability, capabilities, and prices. An alias preserves the accepted name, not the behavior of the older model.
GET /v1/models lists each model once, with its aliases. Looking up an alias with
GET /v1/models/{alias} returns the alias in id and the current model ID in resolved_model.
Completion responses, including streaming chunks, retain the model ID you requested. The
X-Aurora-Resolved-Model response header identifies the current model selected for the request.
Monitor usage and costs
Open Inference API → Dashboard in the Aurora Dashboard to view token usage and costs by model. You can filter the data by time range. Usage through compatibility aliases is grouped under the resolved model, so aliases share one usage series with that model.