OpenWeightsTerminal

Qwen3.8-27B

Qwen3.8-27B

DGX Spark · NVFP4 · TensorFold 0.6.3

Decode · tok/s
42.3
Prefill
—
tok/100W
17.6
Qwen3.8-27B NVFP4 on one GB10 through CapyCTL: TensorFold 0.6.3 / vLLM 0.30.0 / SGLang 0.5.21. tok/s, 1 stream: 42.3 / 28.7 / 31.8

First-person one GB10. Rank is the quoted TensorFold 42.3 single stream.

Officialclaimed
Open pair