Set monthly budget limits and alerts for model inference at the account, workspace, or API Key level, with optional auto-stop when the budget is exhausted.
Overview
Budget management allows you to set monthly budget limits and alerts for model inference, and optionally automatically restrict pay-as-you-go API calls when the budget is exhausted, providing flexible cost control at the account, workspace, or API Key level.
Log in to the Model Studio console. In the left-side navigation pane, choose Cost and Usage > Budget Management to go to the Budget Management page.
You can set budgets independently at the account, workspace, and API Key levels. For details, see Budget dimensions.
Budget scope and statistics cycle
- Budget scope: Only includes pay-as-you-go API charges. Does not include model training, preset throughput (PTU), or model unit (DTU) charges.
- Statistics cycle: Each calendar month, for example, September 1 00:00 to September 30 24:00.
Budget dimensions
You can set budgets independently at the following three dimensions:
Dimension | Scope | Description |
|---|---|---|
Account | All workspaces and API Keys under the primary account | Sets an account-level total budget, aggregating all pay-as-you-go charges under the account. |
Workspace | A single workspace | Sets an independent budget for the specified workspace, counting only pay-as-you-go charges within that workspace. |
API Key | A single API Key | Sets an independent budget for the specified API Key, counting only pay-as-you-go charges generated by that Key. |
Set budget and alerts
On the Budget Management page, click Edit for a record in the table, or click the Set Budget button above the table. In the Set Budget and Alerts panel, configure the following options:
- Budget Target: Shows the dimension information, such as the primary account (all workspaces) and account ID.
- Monthly Budget Management: Enable to set a budget amount.
- Current Budget: When editing, shows the existing budget amount; shows "Not set" for first-time setup.
- New Budget: Enter the monthly budget limit.
- Stop Service When Budget Reached: When enabled, the system automatically restricts pay-as-you-go calls for that dimension when costs reach the budget limit.
- Alert Threshold: Set a percentage of the budget (such as 80%). A notification is sent when costs reach this percentage. The threshold automatically calculates the alert amount based on the budget—for example, with a budget of ¥10,000 and a threshold of 80%, it displays "Send notification when consumption reaches 80% (¥8,000.00) of the budget". Click Set Notification Method to configure notification channels.
Behavior when budget is reached
Budget alerts
When costs reach your configured alert threshold (such as 80% of the budget), as well as when the budget is exhausted (100%), the system sends a notification.
Stop service when budget is reached
If Stop Service When Budget Reached is enabled, the system automatically restricts pay-as-you-go calls for that dimension when costs reach the budget limit. Calls will return a 429 error. Example error message:
The budget configured in Budget Management has been exhausted. You will be unable to make further API calls until the budget limit is increased or reset.For detailed information about this error, see Error codes.
FAQ
Does budget management include model training charges?
Does budget management include model training charges?
No. Budget management only tracks pay-as-you-go API charges, excluding model training, preset throughput (PTU), and model unit (DTU) charges.
How do I resume calls after the budget is reached?
How do I resume calls after the budget is reached?
You can resume calls by:
- Increasing the budget limit on the Budget Management page.
- Turning off the "Stop Service When Budget Reached" toggle.
Why are charges still incurred after reaching the budget limit?
Why are charges still incurred after reaching the budget limit?
Service stop has a certain delay, and charges incurred during the delay will still be collected normally. We recommend monitoring alerts promptly when the alert threshold (such as 80%) is reached to avoid exceeding the budget.