Starting December 9, 2025, Function Compute adds two new billing mechanisms for CPU functions: an hourly minimum charge and a default 1-minute delayed release for on-demand instances. When a function scales in to zero instances, no charges apply.
Background
A small number of users (less than 0.1% of cases) create many small-spec functions and invoke them infrequently, generating almost no compute or storage costs. To maintain stable, high-performance services for all users and prevent extreme usage patterns from affecting platform stability, Function Compute is adjusting its billing model.
Change details
The following two mechanisms take effect on December 9, 2025, at 00:00. Both apply to CPU functions only.
No charges when a function scales in to zero instances. If a function has no invocations and no running instances within an hour, no charge applies.
Hourly minimum charge
If a CPU function is invoked or uses compute resources within an hour, a minimum charge of CNY 0.01 applies for that hour. If your actual usage cost is less than CNY 0.01, you are billed CNY 0.01. Otherwise, you are billed for actual usage.
Default 1-minute delayed release for on-demand instances
This mechanism is being gradually rolled out in a grayscale release starting March 30, 2026.
After a request to a CPU function completes, its on-demand instance is held for a default 1-minute delay before being released. This reduces cold starts and helps maintain service stability. You can extend the delay beyond 1 minute if needed.
During the delay, the system determines whether the instance is active or in shallow hibernation and bills accordingly. Idle time between requests on an instance is billed at the shallow hibernation rate by default.
Billing by instance state
| Instance state | When it occurs | Billing |
|---|---|---|
| Active | Instance is processing a request | Billed at the active rate |
| Shallow hibernation | Instance is idle (between requests or during the delay period) | Billed at the shallow hibernation rate |
| Zero instances | No invocations and no running instances within an hour | No charge |
FAQ
Q: Can I still run my functions at zero cost?
Yes. If your function has no invocations and no running instances within an hour, you are not charged. The pay-as-you-go model is preserved.
Q: Is the 1-minute delayed release applied per request or per instance?
The delay is applied at the instance level. For each CPU function instance, compute resources are released 1 minute after its last request completes. If an instance handles multiple requests, the 1-minute delay starts after the final request finishes. Idle time between requests on the same instance is billed at the shallow hibernation rate by default.