community pool / model answer

Qwen3.6-35B-A3B-MTP-GGUF

chat · unsloth/Qwen3.6-35B-A3B-MTP-GGUF

69.4 tok/s · model median · 36 runs in the model record.

Reference rig	Quantization	Median result	Basis	TTFT (ms)	Peak VRAM	Max context	Engines
GTX 1080 Ti 11GB	4-bit	39.4 tok/s	reported	-	10 GB	128,000	llama.cpp
M5 Max 128GB	4-bit	97 tok/s	reported	-	- GB	2,048	llama.cpp
M5 Pro 24GB	3-bit	80.4 tok/s	reported	54	- GB	65,536	llama.cpp
Radeon AI Pro R9700 32GB	4-bit	56.7 tok/s	reported	0	24.5 GB	2,048	llama.cpp
RTX 3060 12GB	2-bit	39.2 tok/s	measured	3,477	- GB	2,048	llama.cpp
RTX 3060 12GB	4-bit	24.6 tok/s	reported	3,009	- GB	8,320	llama.cpp
RTX 3060 12GB ×2	4-bit	98.9 tok/s	reported	86	- GB	2,048	llama.cpp
RTX 3090 24GB	3-bit	195.1 tok/s	reported	24	20.2 GB	163,840	llama.cpp
RTX 3090 24GB	4-bit	121.6 tok/s	reported	87	23.2 GB	262,144	llama.cpp
RTX 4090 24GB	4-bit	213.9 tok/s	reported	-	- GB	32,768	llama.cpp
RTX 5060 Ti 16GB	3-bit	119.1 tok/s	reported	-	- GB	150,000	llama.cpp
RTX 5060 Ti 16GB	4-bit	36.7 tok/s	reported	-	16 GB	131,072	llama.cpp
RTX 5060 Ti 16GB ×2	4-bit	129.1 tok/s	reported	93	- GB	8,192	llama.cpp
RTX 5070 12GB	4-bit	64.5 tok/s	reported	228	- GB	64,000	llama.cpp
RTX 5070 Ti 16GB ×2	4-bit	141 tok/s	reported	123	29,335 GB	163,840	llama.cpp
RTX 5090 32GB	4-bit	222.9 tok/s	reported	89	28.6 GB	65,536	llama.cpp
RTX 5090 32GB	5-bit	265.4 tok/s	reported	57	- GB	2,048	llama.cpp
RX 6700 10GB	3-bit	10.1 tok/s	reported	2,781	4 GB	4,096	llama.cpp
RX 7900 XTX 24GB ×2	4-bit	109.5 tok/s	reported	183	- GB	2,048	llama.cpp
Ryzen AI Max 395 128GB	4-bit	54.5 tok/s	measured	32,830	24.4 GB	262,144	llama.cpp
Ryzen AI Max 395 64GB	4-bit	60.5 tok/s	measured	1,575	21.5 GB	8,192	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.