Model lifecycle
The statuses a model goes through in the Berget AI catalogue (Preview, Active, Deprecated, and Retired) and what each status means for your integrations
This page explains the policy for releasing new models and retiring older ones from the catalogue. Each model has a status that represents where the model is along its lifecycle.
The four statuses
- Preview: This indicates that a model is available for testing, but may be unreliable. A model in Preview isn't guaranteed to graduate to Active and may be removed without notice. We strongly advise against using models in Preview in production.
- Active: The model is fully supported and recommended for production use. An Active model always passes the Deprecated status before it's retired.
- Deprecated: The model has an end-of-life date after which the model is marked as Retired.
- Retired: The model has been removed from the catalogue. Attempting to use the model results in a
model_not_founderror. Retirement is permanent, so plan migrations during the Deprecated status, not after.
Eval and Stable
Previously, we've used Eval and Stable to reflect the model status. Going forward, we'll be using Preview and Active.
Deprecation policy
| Model status | Deprecation notice |
|---|---|
| Preview models | No notice |
| Active models | 2 or 3 weeks, depending on model usage |
How to check the model lifecycle
To monitor the status for each model, you can query GET /v1/models and parse the following fields on each model:
| Status | Meaning |
|---|---|
lifecycle_status | One of preview, active, and deprecated |
end_of_life | The date when the model is retired. Only included if lifecycle_status is deprecated. |
Retired models
| Retired at | Model |
|---|---|
| 2026-09-11 | openai/gpt-oss-120b |
| 2026-09-11 | meta-llama/Llama-3.3-70B-Instruct |
| 2026-09-04 | zai-org/GLM-4.7-FP8 |
| 2026-08-28 | moonshotai/Kimi-K2.6 |
| 2026-08-25 | mistralai/Mistral-Medium-3.5-128B |
| 2026-06-24 | meta-llama/Llama-3.1-8B-Instruct |