#
Long-Term Support (LTS)
For an LTS model, PHOENIQS provides support for at least six months after we announce its deprecation. The six months begin on the announcement date, not when the model was first released. LTS protects the major-version API identifier during this transition period. It does not mean that we keep the same underlying minor release unchanged.
A new major version may lead us to retire the previous one, but it is not the only reason. We also review models with consistently low usage. Keeping a model available requires dedicated GPU capacity, even when demand is limited. We balance this cost against customer needs so that we can maintain a sustainable service.
#
How minor model updates are introduced
Preview: We make a candidate minor update, such as
GLM 5.4-preview, available for at least one month.Validation: During that month, we check compatibility, model behavior, and operational readiness. We encourage customers to test the preview with their own prompts, integrations, and workflows.
Pricing: Before adopting the update, we confirm that its price is no higher than the current major-version offering.
Promotion: If validation is successful, we update the model behind the existing major-version API identifier. Applications can continue to use the same model name.
Because model responses can vary, a successful validation does not guarantee identical outputs. If your application depends on specific behavior, we recommend testing it against the preview before the update is promoted.
#
Retirement notices and migration
For LTS models, the planned retirement date will be at least six months after we announce deprecation. We publish retirement details in Cloud Docs and notify affected customers where applicable. Notices identify the model, relevant dates, suggested alternatives, and migration guidance when available.
Before final retirement, a model may move to On-Demand status. In that state, active hosting is paused, but customers can request temporary reactivation by opening a service ticket.
Models with an announced retirement date are listed under Scheduled for Retirement. Models that are no longer available are listed under Decommissioned Models.
#
Custom model hosting
If you need a model that is not in our standard catalog, or want to continue using one beyond its planned retirement, contact us about custom hosting. We can review technical feasibility, expected usage, and a consumption commitment with you.
To start, open a service ticket describing the model you need, your expected usage, and your timeline.
#
Discovering and selecting models through the API
Use the model discovery API to see which model identifiers are currently available. A name ending in -preview identifies an experimental model. For production applications, choose a model from the Active Models catalog and use its LTS major-version identifier when one is available.
For support periods, retirement dates, and migration guidance, consult the pages under Model Status and Model Catalog Statuses.