community pool / model answer

Qwen3.5-27B

chat · Qwen/Qwen3.5-27B

25.1 tok/s · model median · 18 runs in the model record.

Reference rig	Quantization	Median result	Basis	TTFT (ms)	Peak VRAM	Max context	Engines
Arc Pro B70 32GB	4-bit	18 tok/s	reported	486	20 GB	8,192	llama.cpp
Arc Pro B70 32GB	5-bit	25.1 tok/s	reported	-	- GB	204,800	llama.cpp
GB10 Grace Blackwell 128GB	4-bit	11.5 tok/s	measured	658	113.5 GB	20,000	vllm
GTX 1080 Ti 11GB	4-bit	2.5 tok/s	reported	-	10.3 GB	8,192	llama.cpp
Radeon 8060S Graphics 103GB	4-bit	104.5 tok/s	reported	338	18 GB	4,096	hipfire
Radeon 8060S Graphics (Strix Halo APU, gfx1151) 96GB	4-bit	14.8 tok/s	reported	480	16 GB	65,536	hipfire
Radeon AI Pro R9700 32GB	4-bit	196.2 tok/s	reported	133	17.9 GB	4,096	hipfire
Radeon AI Pro R9700 34GB	4-bit	286.6 tok/s	reported	111	17.5 GB	4,096	hipfire
RX 7900 XTX 24GB	4-bit	191.5 tok/s	measured	136	17 GB	4,096	hipfire

Context tested

The table above is the tested context for each exact rig and quantization cell. Empty fields remain empty rather than being filled with an estimate.