Pricing
4ALL API uses a prepaid credit system: top up first to get credits, then deductions are made based on actual usage when calls are made. The real-time prices for all models are subject to the console Model Marketplace; this page only explains the billing rules.
Two Billing Methods
| Method | Applicable Models | Rules |
|---|---|---|
| Usage-based billing (by token) | Text-based models such as chat, embeddings, and image recognition | Input and output tokens are priced separately, with output usually higher than input |
| Per-request billing (per request) | Media models such as image generation and video generation | A fixed price is charged for each successful request, regardless of token usage |
A given model will use only one of these methods; the price label for each model in the Model Marketplace indicates whether it is “usage-based” or “per-request”.
Deduction Process
- Pre-authorization: When a request enters, the system pre-deducts a credit amount based on the model’s maximum usage;
- Settlement: After the request completes, the final charge is adjusted based on actual usage;
- Refund on failure: Failed requests (such as upstream errors or timeouts) are not charged, and the pre-deducted credits are automatically returned. Asynchronous tasks such as video jobs are also refunded when the task fails.
Tokens and Groups
- An account can create multiple API tokens, and you can set separate credit limits and expiration times for each token, making it easier to control budgets by project;
- The group that a token belongs to determines the available models and multipliers. For example, the
freegroup can only call free models; - The current balance can be viewed in the console, and can also be retrieved programmatically through the Balance Query API.
Common Questions
- When will the credits arrive after topping up? They are credited immediately after payment succeeds.
- Do credits expire? Account credits do not expire; tokens can be set with an expiration time separately.
- How do I estimate costs? Run a batch of typical requests against the target model, check the actual charges for each call in the console’s “logs”, and then extrapolate based on business volume.
See the Billing and Credits FAQ for more billing details.