Verification checklist
- ✓Open Bailian model settings and check whether “stop when free quota ends” is on. Unverified accounts are forced on; verified accounts default off.
- ✓Confirm Aliyun real-name status. Completing verification does not flip the switch back to stop-on-exhaustion.
- ✓Export bill details. The last field of Instance ID is the stop-on-exhaustion flag. Split by model, ApiKeyID, and channel.
- ✓Confirm the region is China North 2 (Beijing) and the path is realtime inference. Batch, tuning, and deployment cannot draw free quota.
- ✓Token Plan / Coding Plan dedicated keys skip free quota and bill pay-as-you-go. Do not reconcile those lines against remaining free tokens in the console.
1. Real-name verification opens pay-as-you-go, not “still free”
Finance often merges two complaints: “we still have free quota, why a bill?” and “quota is gone, why are we still charged?” Aliyun Bailian’s free-quota page splits them.
Before real-name verification, “stop when free quota ends” is forced on: calls stop at exhaustion and do not convert to pay-as-you-go. After verification, the same switch defaults off. Further calls bill against the Aliyun balance. A short balance becomes arrears; pay-as-you-go traffic can then be paused.
“We already verified” is not “we are still on the free path.” Verification opens the paid path. Treat it as a gift extension and the invoice will not match. Config changes also lag: requests already in flight after you flip stop-on-exhaustion can still bill. A screenshot of the switch and a bill timestamp are not the same second.
2. What free quota covers, and what it does not
Free quota is not a shared token pool for the whole account. Official coverage is narrow: China North 2 (Beijing) realtime inference only. Batch inference, tuning, and deployment cannot draw it. The 90-day window starts from the later of activating Model Studio, the model’s release, or approval of the quota request.
The primary account and RAM users share one pool. Workspaces log separately; charges roll up. Team A can still show leftover quota while Team B’s key has already pushed the account onto pay-as-you-go.
Deduction order is free quota > resource packs > model savings plans > general AI savings plans > pay-as-you-go. Exhausting free quota does not stop traffic; it falls to the next layer. If stop-on-exhaustion (safe mode) is on, service stops and savings plans cannot offset until you turn the switch off. A screenshot of remaining resource packs does not explain a post-free-quota bill—that is the order working as documented.
3. Dedicated keys, snapshots, and latest are separate ledgers
Dedicated API keys for Token Plan and Coding Plan skip free quota and bill pay-as-you-go. The cost overview can still show leftover free tokens; those keys never draw that column. Reconcile by key, not by “account remaining quota.”
Model snapshots and the latest alias have independent free quotas. Traffic on qwen-plus-latest plus a snapshot for regression burns two buckets. One still looks unused while the other has already flipped to pay-as-you-go.
The last field of “Instance ID” on the bill is the stop-on-exhaustion flag. Billing docs state pay-as-you-go uses hold-then-monthly-settle: posting a line is not the same as debiting cash. Minute-level posting lag means bills can still appear after you stop calling. Treat a hold as money already gone, or delayed posting as continued usage, and you will misread the period.
Available credit below zero is arrears. During arrears, leftover free quota, packs, and savings plans will not keep pay-as-you-go calls alive. Coding Plan / Token Plan package quota is independent of cash balance and can still run, but auto-renewal fails. Aliyun credit is account-level: unpaid ECS or OSS can drive the same balance negative and pause Bailian with it.
4. Volcengine defaults overage off; verified Bailian defaults to payg
Volcengine makes the contrast sharper. Agent Plan terms say overage pay-as-you-go defaults off and must be turned on. If the plan is exhausted and overage is off, the capability may stop. If you enable it, overage bills at the matching standard API price. Campaign copy also says exhausted plan quota will not drain other packs or account balance.
The same sentence—“free quota is gone”—means “start pay-as-you-go” on verified Bailian’s default path, and “stop” on Volcengine’s default path. Treat either default as an industry rule and you will overdraw one vendor while filing a false outage on the other.
Error AllocationQuota.FreeTierOnly appears only when stop-on-exhaustion is on and quota is gone. On a verified account with the switch off, you will not see that error; you will see a bill. No error is not no charge.
5. Split the ledger; keep overseas seats off the cloud account
Reconcile at least three columns: free-quota offset, pack or savings-plan offset, and cash pay-as-you-go. Then split region, Batch versus realtime, and dedicated keys. Which column is rising tells you whether to turn stop-on-exhaustion on, retire a dedicated key, or clear ECS arrears first.
Bailian settles against the Aliyun balance and domestic invoicing, not an international card rail. Do not mix overseas seat renewals and cloud top-ups into one “AI was expensive this month” card. Isolate web-seat renewals on a capped virtual card such as RDVCC so those auto-renewals do not compete with Aliyun top-ups. A virtual card does not change Bailian’s stop-on-exhaustion default—the switch is in the console, not in the BIN.