Model Lifecycle and Retirement
Models may be retired as newer versions or higher-performing alternatives become available.
This page describes the retirement policy and the customer actions required before a model is retired.
Model Lifecycle Overview
Models are periodically retired as part of the normal service lifecycle.
Retirement may happen when:
- Newer model versions become available.
- Newer models surpass existing models in capability, performance, efficiency, or overall quality.
Retirement Notifications
Customers are notified before planned model retirement:
| Notice period | Guidance |
|---|---|
| 30 days before retirement | Notification is sent before the scheduled retirement date. |
| 14 days before retirement | A second notification is sent before the scheduled retirement date. |
Replacement Models
At least one alternative model is made available before retirement to support continuity and switchover.
Migration Requirements
Customers are responsible for updating applications, integrations, and API calls before the retirement date.
Migration actions include:
- Replacing retired model identifiers in requests.
- Updating associated API keys or configuration settings as needed.
Retired models
After a model is retired, it no longer accepts inference requests.
Support and Guidance
For questions or guidance, contact the designated Service Delivery Manager.