Discover · Models
Compare prepared models for the work you need to do
SAVRN Cloud is preparing two specific Qwen configurations for SAVRN-operated inference. Start with the intended task, then inspect the artifact and the evidence still needed before release. The public comparison helps you prepare a request; it does not make either candidate callable. Connected execution is Coming soon.
The first SAVRN model services
Models in preparation
Qwen3.5-9B
Utility candidate
Prepare supplied-evidence review or structured extraction. Task fit remains to be evaluated; connected execution is not available.
- Proposed artifact
- Q6_K
- Live model access
- Pending qualification
Qwen3.8-27B
Coding candidate
Prepare research-code assistance with explicit assumptions and checks. Generated code is not executed, and model qualification is pending.
- Proposed artifact
- Q5_K_M
- Live model access
- Pending qualification
Qwen3.5-9B Q6_K: utility candidate
The utility candidate is a Q6_K artifact based on Qwen3.5-9B. Its intended starting tasks include reviewing supplied public evidence and extracting fields from supplied text. Those roles are a proposed evaluation scope, not a claim that the candidate has passed a benchmark. The canonical Model Hub entry describes the base model; the Cloud profile describes this prepared configuration.
Qwen3.8-27B Q5_K_M: coding candidate
The coding candidate is a Q5_K_M artifact based on Qwen3.8-27B. Research code is its suggested starting workflow: draft an explanation or a small code change, identify assumptions and propose checks a researcher could run. The preview neither invokes this model nor executes generated code. A larger model label does not establish better results for a particular task.
Compare the same task under explicit limits
Both candidates currently use proposed pilot limits of 8,192 total context tokens and 1,024 output tokens. These are intended bounds for future qualification, not measured service guarantees or the publishers’ maximum context claims. Compare task fit, precision, input requirements and missing evidence first. There are no published SAVRN quality scores, throughput measurements, prices or available customer capacity for these candidates.
Keep preparation distinct from demonstration
Choose Prepare request from a candidate profile to carry its identity into a guided playground sample. A separate fictional fixture supplies the deterministic browser response. The receipt distinguishes the intended candidate from that actual demonstration fixture. Existing fictional Research 8B, Code 14B and Reasoning 32B profiles remain workflow examples; they are not alternate names for Qwen models.
Understand the release decision
A prepared artifact must still pass identity and provenance review, task-specific evaluations, supported-feature tests and performance measurement on independently allocated capacity. License eligibility, project access, pricing and usage accounting need separate approval. Independently allocated serving capacity remains to be approved before customer access can be released.
Service maturity
Explore the workflow today
Two prepared Qwen candidate profiles, a comparison and guided request preparation are available alongside clearly separated fictional fixtures.
Before connected service access
Neither Qwen candidate is available for connected execution. Qualification, independent serving capacity and an explicit service release are pending.
What this makes possible
The intended outcome is a defensible model choice tied to the task, supported configuration and evidence available at the time of selection.
Build toward the work that matters.
Tell SAVRN what your institution needs to run, who reviews the results and where its data must stay. That workload defines the next service to qualify.
Discuss an AI project