跳转到主要内容
2026/10/27 09:25 (UTC+8) · Model Status ⚠️ The October 26 incident review and compensation details have been published. We sincerely apologize to all customers affected by the service interruption and the instability following our emergency migration. What happened On Monday, October 26, we observed a decline in request volume starting at 09:48. Core API service was restored at 11:10 (Beijing time, UTC+8), making the main incident window 82 minutes. Some issues caused by the emergency migration continued to affect customers afterward. This was an unacceptable service interruption. The incident was triggered by a faulty connection between the origin server’s disk and motherboard. Data and billing Our database is hosted separately on RDS with another cloud provider. Customer data was unaffected. Requests that did not reach our platform during the outage were not charged. Subsequent issues and an IP allowlist reminder Following the emergency migration, some customers also experienced:
  • Some models becoming unavailable;
  • Frequent 429 responses;
  • 401 responses from previously working API tokens with IP allowlists configured.
We use CDN acceleration and reverse proxies to improve access speeds from mainland China. During the migration, these mechanisms affected IP allowlist validation, causing some tokens to return 401. These subsequent issues also disrupted customers’ use of our service, and we take responsibility for them. If you temporarily removed your IP allowlist because of 401 errors, please restore the original configuration promptly to reinstate access restrictions and protect your token. Remediation and compensation This incident exposed gaps in our emergency preparedness: we did not have a ready-to-start cold standby server with a fully prepared runtime environment, and our rapid failover capabilities were inadequate. Maintaining service continuity is our responsibility, and we fell short. Our previous P0 incident occurred on November 8, 2025. We will conduct a thorough review, prepare the backup environment, improve our failover capabilities, and validate these improvements through recovery drills. As an apology, APIYI will automatically apply a 10% discount to usage in both the Default and SVIP groups throughout Wednesday, October 28, 2026 (Beijing time, UTC+8). No application is required. These two commonly used groups cover most models. Models available in both groups have identical pricing, while SVIP also provides access to some less commonly available models. We chose a proportional discount based on actual usage that day, rather than a fixed credit per account. This ties the amount saved to usage and provides larger reductions for customers with higher usage. Thank you for your understanding, patience, and continued support. We will turn the lessons from this incident into concrete improvements and work to earn your trust through reliable, long-term service.
← Back to Live Updates · 📚 Monthly Archive