# Models, offers and deployments

Canonical: https://savrn.com/cloud/docs/concepts-models-offers-deployments

SAVRN Cloud · Coming soon. Public documentation and a browser simulation are available; connected services are not yet available.

A Qwen base-model record describes discovery evidence; a Cloud candidate describes a specific artifact awaiting qualification. Follow that identity through intended service limits, allocated deployment and request receipts, keeping the actual fictional fixture visible whenever the public preview produces a sample.

**SAVRN Cloud · Coming soon · Public preview · Browser-local simulation · No live compute or payments**

## Keep each identity explicit

A base model, a prepared artifact and a callable service answer different questions. The [SAVRN Model Hub](/models) describes source-backed model identities. The Cloud [candidate comparison](https://savrn.com/cloud/console/#/models/compare) describes two prepared configurations: Qwen3.5-9B Q6_K and Qwen3.8-27B Q5_K_M. Neither profile is a connected offer or a capacity reservation.

| Record | Question it answers |
|---|---|
| Catalog entry | Which base model and published evidence are described? |
| Prepared candidate | Which artifact and intended workload are being considered? |
| Qualified configuration | Which exact runtime, limits and features passed review? |
| Offer revision | What can an authorized project use, under which approved terms? |
| Deployment | Which allocated resource serves that configuration? |
| Request and receipt | What actually ran and what usage was recorded? |

## Trace a sample without confusing it with execution

Prepare a request from either candidate profile, choose a guided starter and inspect the [Playground](https://savrn.com/cloud/console/#/playground). The intended candidate remains attached to the prepared task. A separate fictional model fixture produces the deterministic browser result. The receipt must identify that fixture as the actual demonstration source rather than attributing the output to Qwen.

Follow the record to [Usage](https://savrn.com/cloud/console/#/usage) and compare it with the active [Project](https://savrn.com/cloud/console/#/projects). Local identifiers help connect the example, but a matching name alone does not establish artifact identity. Exported records should retain their simulation labels.

## Make qualification specific

The proposed pilot bounds of 8,192 context tokens and 1,024 output tokens are future test limits, not publisher maxima or measured guarantees. Changing precision, serving software, hardware, configuration or supported features can change behavior and requires renewed evidence. A prepared weight file does not establish task quality or spare capacity.

## Preserve the release boundary

Connected execution is **Coming soon**. Independently allocated SAVRN-operated capacity, access controls, reviewed license eligibility and approved pricing remain separate requirements. Sample Work Credits do not supply those approvals. Public documentation should expose useful model identity and readiness information without exposing credentials, private endpoint addresses or another project's records.