Configs › Qwen 3.8 27B on Strix Halo, DGX Spark and Mac: measured tok/s

Qwen 3.8 27B on Strix Halo, DGX Spark and Mac: measured tok/s

Every tracked benchmark config for Qwen 3.8 27B on local hardware: quant, backend, context, tokens per second, source-linked and trust-tiered.

Snapshot 2026-09-16 · 158 configs · 75 models. Every number is generated from the tracker snapshot and linked to where it was measured; nothing here is typed by hand.

Measured configs for Qwen 3.8 27B

Model	Quant	Backend	Decode	Context	Trust	Source
Qwen 3.8 27B	FP4		33.8 tok/s		Owner-submitted	source
Qwen 3.8 27B	FP4		26.9 tok/s	32768	Owner-submitted	source
Qwen 3.8 27B	FP8	llama.cpp	19 tok/s		Owner-submitted	source
Qwen 3.8 27B	Q4_K_M		12.4 tok/s		Owner-submitted	source
Qwen 3.8 27B	Q4_K_M	llama.cpp	12.3 tok/s	32768	Owner-submitted	source
Qwen 3.8 27B	FP16	llama.cpp	5 tok/s		Owner-submitted	source

More: all configs · hardware · methodology · llms.txt · llms-full.txt · API (OpenAPI) · JSON snapshot. Built by Altronis, private on-prem AI, Singapore.