Skip to content

Preview models of AI Model Serving

Last updated on

Preview Models are models that are made available in STACKIT AI Model Serving before they are fully established in the model portfolio. They give you early access to new models so you can evaluate them and provide feedback.

Preview Models are always prefixed with public-preview-- in their model name. They are also marked with the Public Preview label in the Available Shared Models overview.

How Preview Models differ from regular Shared Models

Section titled “How Preview Models differ from regular Shared Models”

Preview Models come with reduced guarantees compared to regular Shared Models:

  • No SLAs: Preview Models are not covered by Service Level Agreements. The availability and performance guarantees of regular Shared Models do not apply.
  • No deprecation phase: Preview Models can be removed from the service at any time without a deprecation phase or prior transition period.
  • Rate limits: The TPM and RPM limits of Preview Models may differ from those of regular Shared Models. Requests to increase rate limits are evaluated on a case-by-case basis.
  • Support: Support requests for Preview Models are handled on a best-effort basis.
  • Data usage: Your data is not used for any purpose beyond serving your requests, the same as for regular Shared Models.

A Preview Model can be promoted to a generally available (GA) model after evaluation, but this is not guaranteed. Models that do not meet the requirements for GA may be removed from the service at any time.

You can provide feedback on Preview Models, for example on model capabilities or behavior, by opening a support ticket.

Preview Models should not be used for production workloads. Use them for:

  • Evaluating new models for potential adoption
  • Testing and prototyping
  • Providing feedback on model capabilities and behavior

For production workloads, use models with a Supported status.