community pool / model answer

Qwen3.6-35B-A3B-GGUF

chat · unsloth/Qwen3.6-35B-A3B-GGUF

56.9 tok/s · model median · 53 runs in the model record.

Reference rig	Quantization	Median result	Basis	TTFT (ms)	Peak VRAM	Max context	Engines
CMP 170HX 64GB	8-bit	104.9 tok/s	measured	328	- GB	16,384	llama.cpp
M4 Pro 48GB	4-bit	47.1 tok/s	reported	-	- GB	640	llama.cpp
M5 32GB	4-bit	34.3 tok/s	reported	-	- GB	64,000	llama.cpp
Radeon 890M 32GB	4-bit	20.1 tok/s	reported	1,530	- GB	2,048	llama.cpp
Radeon AI Pro R9700 32GB	4-bit	78.6 tok/s	reported	178	- GB	2,048	llama.cpp
Radeon RX 7900 XTX 24GB	4-bit	57.6 tok/s	reported	-	19.5 GB	640	llama.cpp
Radeon RX 9070 16GB	3-bit	76 tok/s	reported	360	15.5 GB	65,536	llama.cpp
RTX 3060 12GB	2-bit	75.3 tok/s	reported	63	- GB	65,536	llama.cpp
RTX 3060 12GB ×2	4-bit	78.2 tok/s	measured	224	21.6 GB	98,304	llama.cpp
RTX 3090 24GB ×2	3-bit	125.9 tok/s	reported	-	10.1 GB	131,072	llama.cpp
RTX 3090 24GB ×2	4-bit	124 tok/s	reported	-	22.5 GB	1,048,576	llama.cpp
RTX 3090 24GB ×2	6-bit	40.3 tok/s	reported	463	19.8 GB	32,768	llama.cpp
RTX 4070 12GB	4-bit	28.4 tok/s	reported	772	- GB	2,048	vllm
RTX 4090 24GB	4-bit	165.1 tok/s	reported	-	23,691 GB	64,000	llama.cpp
RTX 5060 Ti 16GB ×2	4-bit	94.4 tok/s	reported	81	- GB	8,192	llama.cpp
Ryzen AI Max 385 24GB	4-bit	19.2 tok/s	reported	1,530	- GB	2,048	llama.cpp
Ryzen AI Max 395 64GB	4-bit	60.5 tok/s	reported	963	21.5 GB	8,192	llama.cpp
Tesla P100 32GB ×2	16-bit	67 tok/s	measured	3,541	- GB	2,048	llama.cpp
Tesla P100-PCIE-16GB 16GB	3-bit	24.2 tok/s	reported	-	- GB	2,048	llama.cpp
Tesla P100-PCIE-16GB 16GB ×2	5-bit	63.9 tok/s	reported	-	- GB	2,048	llama.cpp
Tesla V100 32GB	4-bit	58.3 tok/s	measured	71,447	21.5 GB	262,144	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.