Dedicated Inference billing model and prices
Balance
To pay for Cloud Platform resources, depending on the balance type in the account, the unified balance or the Cloud Platform balance is used.
You can pay for resources using various types of funds: main funds, bonuses.
Before paying, top up the balance.
Billing model
The Cloud Platform uses the pay-as-you-go billing model. Every hour, funds are deducted from the balance for the previous hour of using Cloud Platform resources.
Billing for inference service resources is calculated per project. Each project contains a resource group: GPU type and quantity.
Every clock hour, the cost of the resource group is updated.
All Cloud Platform resources for Dedicated Inference are quota-based. For quota-based resources, the maximum consumption during an hour is taken into account.
For example, at 13:25, an inference service was created with two inference instances. Configuration of each inference instance: one NVIDIA® A30 GPU with 24 GB of memory. At 13:40, the inference service was scaled down — the number of inference instances was decreased to one. The bill for the hour 13:00–14:00 will include the consumption of two NVIDIA® A30 GPUs with 24 GB of memory. For the hour 14:00–15:00, only the consumption of one NVIDIA® A30 GPU with 24 GB of memory will be taken into account, unless the number of inference instances was increased or other inference services were created.
If the number of inference services or the number of inference instances in an inference service increases in a project, the resource charges will change immediately.
For example, at 13:25, an inference service was created with one inference instance. Configuration of the inference instance: one NVIDIA® A30 GPU with 24 GB of memory. At 13:40, the inference service was scaled up — the number of inference instances was increased from one to two. For the hour 13:00–14:00, funds will be deducted for two NVIDIA® A30 GPUs with 24 GB of memory.
Resource blocking if there are insufficient funds on the balance
If at the time of debiting there are insufficient funds on the balance for payment, inference service operations will be suspended, GPUs of the created inference services will be released to the Cloud Platform shared resource pool, and billing will stop. Resources from the shared pool can be used by other clients.
To resume inference service operations, top up the balance by the debt amount within 8 hours of suspension. If the balance is not topped up by the debt amount within this period, all created inference services will be permanently deleted. Projects will not be deleted.
To make sure that there are always sufficient funds on the balance, you can configure balance notifications and automatic balance top-ups.
Prices
The cost depends only on the type and number of GPUs in the inference service configuration. The number of tokens does not affect the cost.
GPU prices can be viewed when creating an inference service in the Control panel.
Accounting documents
After payment, you can obtain accounting documents.