The lifecycle
- 01
Preview
A new model can launch here. Its parameters, limits and price can still change, and it can be withdrawn instead of moving to GA. Evaluate it, and keep another model behind production features.
- 02
General Availability
The model is ready for production traffic. Most models are here.
- 03
End of Life
The model is scheduled for removal. It still answers, so you have time to move your traffic.
- 04
Removed
The model is no longer offered. Requests that still name it are refused.
Notice before removal
Our terms commit to at least 14 days of notice, as far as possible, before we remove a model you use. When a model you call reaches End of Life, switch the model field of your requests to another model from the catalogue before it is removed.
Dedicated endpoints
An endpoint you deployed keeps running while its model is in End of Life, and after the model is removed. Once the model is removed, you can still change the endpoint’s replica bounds or delete it. You can no longer redeploy it, change its capacity mode, scaling policy or engine settings, or create a new endpoint on that model.
Move to another model by creating a new endpoint on it, switching your traffic, then deleting the old endpoint.
Where the stage shows
In the console, use the Availability filter on the Models page. Preview and End of Life models show the icon of their stage after their name, the same icons as above.
In the API, GET /api/models returns availability on each model: preview, ga or eol.