# Build research workflows you can understand and govern

Canonical: https://savrn.com/cloud/

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

SAVRN Cloud brings model selection, research work and result review into one workspace on savrn.com. Two SAVRN-operated configurations are in preparation: Qwen3.5-9B Q6_K for utility work and Qwen3.8-27B Q5_K_M for coding tasks. Explore their intended roles and prepare a request in the public console. Connected execution is Coming soon; the preview returns clearly labeled browser-local samples.


## Begin with the evidence you need

A useful research task specifies its sources, the question being asked and the form of an acceptable answer. SAVRN Cloud offers three guided starting points: Evidence review, Structured extraction and Research code. They help a team define that contract before evaluating a model. Use supplied public or invented material; a link in a prompt does not cause the preview to fetch a source.

## Compare two prepared configurations

Qwen3.5-9B Q6_K is the utility candidate; Qwen3.8-27B Q5_K_M is the coding candidate. These are descriptions of specific prepared artifacts, not measured quality rankings or available commercial offers. Their proposed pilot limits are 8,192 total context tokens and 1,024 output tokens. Separate serving capacity, qualification and access approval remain pending.

## Keep a request connected to its project

The console connects a prepared request to its selected project, intended candidate, demonstration fixture and sample usage receipt. A candidate profile supplies the intended model context; a fictional fixture supplies the browser demonstration. That distinction stays with the result so a reviewer can tell what was selected and what actually happened. Sample Work Credits illustrate recordkeeping, without publishing a customer price.

## Build toward SAVRN-operated execution

The service direction is to run qualified model configurations on compute SAVRN operates, whether owned or specifically rented. A connected release needs an independently allocated serving lane, verified identity, task evaluations, measured performance, enforceable access and reconciled usage. Hardware changes and new quantizations require their own evidence.

## Continue from a sample into a review

After exploring a starter, inspect its request receipt and identify what a real evaluation would need to establish. Projects, datasets, evaluation plans and Work Orders provide places to organize that next step. The public preview remains a browser-local workspace: it does not call a model, run supplied code, allocate GPUs or take payment.

## What you can explore today

Compare two prepared SAVRN model candidates and explore three guided starters through labeled browser-local samples.

## Before this service launches

Connected execution requires independently allocated SAVRN-operated capacity, completed qualification, approved access and validated accounting.

## The customer outcome

The intended outcome is a research workflow whose scope, execution choices and review evidence remain understandable as teams and infrastructure change.

## Related documentation

- [Welcome to SAVRN Cloud](https://savrn.com/cloud/docs/start-welcome)
- [First inference walkthrough](https://savrn.com/cloud/docs/start-first-api-request)
- [What is available](https://savrn.com/cloud/docs/start-service-status)
- [Compare prepared models](https://savrn.com/cloud/console/#/models/compare): Review intended roles, artifact precision and qualification gaps.
- [Explore guided starters](https://savrn.com/cloud/console/#/playground): Prepare a research request and inspect a labeled sample.
- [Browse the Model Hub](https://savrn.com/models): Read the source-backed model discovery catalog.