What happened
During a routine billing audit, we found that a small number of users were over-charged for cache tokens between 2026-04-08 and 2026-04-10 due to a temporary issue on an upstream provider's usage API.
Scope
- Time window: 2026-04-08 00:00 to 2026-04-10 23:59 (UTC+8)
- Affected requests: about 0.3% of Anthropic model calls
- Overcharge per call: typically between ¥0.01 and ¥0.50
Compensation
We have completed a full audit. All over-charged amounts will be automatically refunded to your account balance within 24 hours.
- No manual action required
- Refunds appear as
refund-cache-correctionentries in billing details - If your total refund exceeds ¥10 you will also receive a WeChat/email notification
What we are doing about it
- We are upgrading our secondary reconciliation pipeline for cache tokens (see Understanding Cache Token Pricing)
- Every upstream response now passes through an independent verification channel to catch rare upstream anomalies
We apologize for the inconvenience. Questions? Contact support@tokensmart.ai.