community pool / model answer

Qwen3.6-27B-GGUF

chat · unsloth/Qwen3.6-27B-GGUF

24.8 tok/s · model median · 50 runs in the model record.

Reference rig	Quantization	Median result	Basis	TTFT (ms)	Peak VRAM	Max context	Engines
Arc Pro B70 32GB	4-bit	26 tok/s	measured	-	- GB	2,048	llama.cpp
M4 Pro 48GB	4-bit	12 tok/s	reported	-	- GB	640	llama.cpp
M5 Pro 64GB	4-bit	11.4 tok/s	reported	911	- GB	2,048	ollama
Radeon AI Pro R9700 32GB	4-bit	30.2 tok/s	reported	-	- GB	65,536	llama.cpp
Radeon RX 7900 XTX 24GB	3-bit	33.3 tok/s	reported	-	13.5 GB	640	llama.cpp
Radeon RX 7900 XTX 24GB	4-bit	28.5 tok/s	reported	-	14.8 GB	640	llama.cpp
RTX 3060 12GB ×2	4-bit	18.4 tok/s	reported	490	- GB	32,768	llama.cpp
RTX 3090 24GB	4-bit	150.1 tok/s	reported	-	19.5 GB	2,048	llama.cpp
RTX 3090 24GB ×2	4-bit	37.8 tok/s	measured	1,416	12.2 GB	65,536	llama.cpp
RTX 3090 24GB ×2	6-bit	22.2 tok/s	reported	1,532	17.9 GB	32,768	llama.cpp
RTX 3090 24GB ×2	8-bit	23 tok/s	reported	1,659	22.1 GB	32,768	llama.cpp
RTX 3090 Ti 24GB	4-bit	35.7 tok/s	measured	374	21 GB	73,728	llama.cpp
RTX 4070 12GB	2-bit	21.1 tok/s	reported	1,391	11.9 GB	98,304	llama.cpp
RTX 4090 24GB	4-bit	41.8 tok/s	reported	-	- GB	32,768	llama.cpp
RTX 5060 Ti 16GB ×2	4-bit	22.3 tok/s	reported	177	- GB	8,192	llama.cpp
RTX 5090 32GB	3-bit	73.1 tok/s	measured	466	18.4 GB	131,072	llama.cpp
RTX 5090 32GB	4-bit	66.8 tok/s	measured	456	19.5 GB	262,144	llama.cpp
RTX 5090 32GB	5-bit	60.4 tok/s	measured	476	21.7 GB	131,072	llama.cpp
RTX 5090 32GB	6-bit	50.9 tok/s	measured	523	24.6 GB	65,536	llama.cpp
RTX A5000 24GB	4-bit	32.3 tok/s	reported	-	16.2 GB	4,096	llama.cpp
RX 7900 XTX 24GB	3-bit	31.7 tok/s	reported	-	11.6 GB	512	llama.cpp
Tesla P100 32GB ×2	16-bit	17.2 tok/s	measured	10,079	- GB	2,048	llama.cpp

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.