Qwen 3.8 27B
Alibaba's Qwen 3.8 27B. A dense 27 billion parameter model you can teach a custom behavior.
License and lineage
Apache License 2.0
View license Qwen/Qwen3.8-27B
Makojima reviews the exact base revision before accepting a job. The successful merged artifact is downloadable by its owner only when the effective license permits delivery and redistribution.
Reports and independent evidence
Official model card for the reviewed repositoryQualified variants and quality
| Quant | Context | Measured quality change | Status |
|---|---|---|---|
| BF16 | 262,144 | Not measured | Qualification required |
| 8-bit | 262,144 | Not measured | Qualification required |
| 4-bit | 262,144 | Not measured | Qualification required |
A lower-bit option is shown as qualified only after Makojima measures that exact artifact. No quality score is inferred from another model or quant.
Hosting options
Rent a qualified profile by the minute. Billing starts after health checks and stops at shutdown or the spending limit.
Offer Chat, supported Agent actions, and API access without exposing weights. Token prices use one immutable rate card per admitted request.
Compatible GPU classes, throughput, concurrency, and credits per minute appear after serving qualification.