Best home rig for Qwen3.6 35B-A3B

Will it fit my GPU? Best value GPU per euro →

Real, measured generation speeds for Qwen3.6 35B-A3B on home hardware, fastest first. Different quants/engines aren't directly comparable.

Build	Quant	Engine	tok/s	Cost
MacBook Pro M5 Max 128 GB: Qwen3.6 35B-A3B at 271 tok/s (mlx-serve + MTP), Gemma 4 26B-A4B at 166	4-bit MLX (mlx-serve build)	mlx-serve 26.10.2 + MTP	271		source
RTX 5090: 2026 models (llamabench community)	llama-bench tg128 (quant per result)	llama.cpp	258		source
MacBook Pro M5 Max 128 GB: Qwen3.6 35B-A3B at 271 tok/s (mlx-serve + MTP), Gemma 4 26B-A4B at 166	4-bit MLX	Rapid-MLX 0.16.1 + MTP	240		source
4x RTX 3090 lab: Qwen3.6 on vLLM vs llama.cpp	AWQ-INT4 + MTP-3 (TP=2)	vLLM	225-231		source
Mac Studio M5 Max 36 GB: Qwen3.6 35B-A3B at 173 tok/s (Rapid-MLX), 146 on oMLX	4-bit MLX	Rapid-MLX 0.16.1 (MTP)	173.4		source
RTX 4090 (llm-speed signed runs, Oct 2026): Qwen3.6 35B-A3B 161 tok/s, Gemma 4 12B QAT 98 tok/s	Q4_K_M	llama.cpp b8931	160.6		source
4x RTX 3090 lab: Qwen3.6 on vLLM vs llama.cpp	AWQ-INT4 (TP=2)	vLLM	149		source
RTX 3090: 2026 models (llamabench community)	llama-bench tg128 (quant per result)	llama.cpp	146-153		source
Mac Studio M5 Max 36 GB: Qwen3.6 35B-A3B at 173 tok/s (Rapid-MLX), 146 on oMLX	4-bit MLX	oMLX 0.7.0	145.8		source
Single RTX 3090: Qwen3.6 35B-A3B (Giles Thomas)	Q4 GGUF, all layers on GPU, ~90K ctx	llama.cpp CUDA	139.6		source
Radeon RX 7900 XTX (24 GB): Qwen3.6 35B-A3B up to 245K context, with MTP	UD-Q4_K_XL + MTP (draft-n-max 4), ~16K context	llama.cpp ROCm (llama-server)	135.2		source
Mac Studio M5 Max 36 GB: Qwen3.6 35B-A3B at 173 tok/s (Rapid-MLX), 146 on oMLX	4-bit MLX	LM Studio (MLX)	124.5		source
Single Intel Arc Pro B70 with vLLM XPU cookbook	GPTQ Int4	vLLM XPU b70 image, no MTP	120		source
Radeon AI Pro R9700 (one card): Qwen3.6 35B-A3B from 4K to 256K context (AMDai lab rig)	Q4_K_M	llama.cpp Vulkan	117.69		source
4x RTX 3090 lab: Qwen3.6 on vLLM vs llama.cpp	IQ4_XS (1 GPU)	llama.cpp	115-133		source
DGX Spark (GB10): Qwen3.6 35B-A3B NVFP4 at 113 tok/s on vLLM with MTP	NVFP4	vLLM 0.30 + MTP	113	$3,999 (≈€3,568)	source
Radeon AI Pro R9700 (one card): Qwen3.6 35B-A3B from 4K to 256K context (AMDai lab rig)	Q5_K_M (24.68 GB VRAM)	llama.cpp Vulkan	109.73		source
Radeon RX 7900 XTX (24 GB): Qwen3.6 35B-A3B up to 245K context, with MTP	UD-Q4_K_XL + MTP, ~64K context	llama.cpp ROCm (llama-server)	108.2		source
MacBook Pro M5 Max 128 GB: Qwen3.6 35B-A3B at 271 tok/s (mlx-serve + MTP), Gemma 4 26B-A4B at 166	4-bit MLX	mlx_lm.server 0.32.2	107		source
Radeon AI Pro R9700 (one card): Qwen3.6 35B-A3B from 4K to 256K context (AMDai lab rig)	Unsloth UD-Q5_K_XL, short context, f16 KV	llama.cpp Vulkan	102.39		source
DGX Spark (GB10, 128 GB): 23 models on vLLM at 128K context, Qwen3.6 35B-A3B up to 102 tok/s	NVFP4 + MTP (k=2)	vLLM 0.31.0, 128K ctx, 1 user	102.3		source
Minisforum MS-S1 MAX (Strix Halo, 128 GB): Qwen3.6 35B-A3B at 62 tok/s on llama.cpp Vulkan	Q4_K_M (Unsloth)	llama.cpp Vulkan + MTP	93	$3,799 (≈€3,390)	source
Radeon AI Pro R9700 (one card): Qwen3.6 35B-A3B from 4K to 256K context (AMDai lab rig)	UD-Q5_K_XL, 32K context	llama.cpp Vulkan	90.52		source
HP OmniBook Ultra 16 (RTX Spark N1X, 128GB) laptop: local AI benchmarks	Ollama default, 64K context, Codex CLI coding agent	Ollama	89.5	$6,230 (≈€5,559)	source
Radeon RX 7900 XTX (24 GB): Qwen3.6 35B-A3B up to 245K context, with MTP	UD-Q4_K_XL, llama-bench 2K, f16 KV	llama.cpp ROCm	87.2		source
DGX Spark (GB10): Qwen3.6 35B-A3B NVFP4 at 113 tok/s on vLLM with MTP	NVFP4	vLLM 0.30 + MTP	86	$3,999 (≈€3,568)	source
Radeon RX 7900 XTX (24 GB): Qwen3.6 35B-A3B up to 245K context, with MTP	UD-Q4_K_XL, ~16K real-text context, q4_0 KV	llama.cpp ROCm (llama-server)	79.6		source
DGX Spark (GB10, 128 GB): 23 models on vLLM at 128K context, Qwen3.6 35B-A3B up to 102 tok/s	NVFP4, no speculative decoding	vLLM 0.31.0, 128K ctx, 1 user	78.1		source
Radeon AI Pro R9700 (one card): Qwen3.6 35B-A3B from 4K to 256K context (AMDai lab rig)	UD-Q5_K_XL, 128K context	llama.cpp Vulkan	66.43		source
Radeon RX 7900 XTX (24 GB): Qwen3.6 35B-A3B up to 245K context, with MTP	UD-Q4_K_XL, ~64K context	llama.cpp ROCm (llama-server)	66.0		source
Minisforum MS-S1 MAX (Strix Halo, 128 GB): Qwen3.6 35B-A3B at 62 tok/s on llama.cpp Vulkan	Q4_K_M (Unsloth)	llama.cpp Vulkan	62	$3,799 (≈€3,390)	source
Strix Halo (Corsair AI Workstation 300, 128 GB): ROCm vs Vulkan on Qwen3.8 27B and Qwen3.6 35B-A3B	Q8_0 + MTP (draft-n-max 3), coding workload	llama.cpp ROCm	60.92		source
HP ZBook Ultra G1a (Strix Halo laptop, 128 GB): Qwen3.6 35B-A3B at 59.4 tok/s, gpt-oss 120B at 44.7	Q4_K_M	llama.cpp b11213 Vulkan	59.4		source
HP ZBook Ultra G1a (Strix Halo laptop, 128 GB): Qwen3.6 35B-A3B at 59.4 tok/s, gpt-oss 120B at 44.7	4-bit	Ollama 0.34.0	54.4		source
Strix Halo (Corsair AI Workstation 300, 128 GB): ROCm vs Vulkan on Qwen3.8 27B and Qwen3.6 35B-A3B	Q8_0, llama-bench tg128	llama.cpp Vulkan	54.22		source
Minisforum MS-S1 MAX (Strix Halo, 128 GB): Qwen3.6 35B-A3B at 62 tok/s on llama.cpp Vulkan	Q8_0	llama.cpp Vulkan	54	$3,799 (≈€3,390)	source
Minisforum MS-S1 MAX (Strix Halo, 128 GB): Qwen3.6 35B-A3B at 62 tok/s on llama.cpp Vulkan	Q4_K_M (Unsloth)	llama.cpp ROCm	51	$3,799 (≈€3,390)	source
Strix Halo (Radeon 8060S, 96 GB): 2026 models (llamabench community)	llama-bench tg128 (quant per result)	llama.cpp	50.2-62.4		source
Strix Halo (Corsair AI Workstation 300, 128 GB): ROCm vs Vulkan on Qwen3.8 27B and Qwen3.6 35B-A3B	Q8_0, llama-bench tg128	llama.cpp ROCm	47.47		source
Radeon AI Pro R9700 (one card): Qwen3.6 35B-A3B from 4K to 256K context (AMDai lab rig)	UD-Q5_K_XL, 256K context	llama.cpp Vulkan	44.4		source
Radeon RX 7900 XTX (24 GB): Qwen3.6 35B-A3B up to 245K context, with MTP	UD-Q4_K_XL, ~245K context, q4_0 KV	llama.cpp ROCm (llama-server)	38.3		source

Rent before you buy: try this model on a cloud GPU for a few dollars first: Runpod (referral link; sign up with Google to get the new-user bonus)

Compare: DGX Spark vs Mac Studio vs Strix Halo: measured tok/s and tok/s per watt

Memory math: how much VRAM / RAM does Qwen3.6 35B-A3B need?

Qwen3.6 35B-A3B has 35.51B parameters, about 3B active per token (source: GGUF param count via llamabench.ai; 3B active per name). Weight size = parameters × bits-per-weight ÷ 8, before context (KV cache) and runtime overhead.

Format	Bits/weight	Weights	Smallest common memory size that holds the weights alone
Q8_0	8.50	37.7 GB	48 GB
Q6_K	6.56	29.1 GB	32 GB
Q5_K_M	5.70	25.3 GB	32 GB
Q4_K_M	4.89	21.7 GB	24 GB
Q3_K_M	4.00	17.7 GB	24 GB
Q2_K	3.16	14.0 GB	16 GB

Bits-per-weight for GGUF quants are llama.cpp's own figures (llama.cpp quantize README). "As measured" rows apply the source's stated bits-per-weight to the parameter count; real files differ slightly because some layers stay at higher precision. Try other combinations in the VRAM calculator.

Why it can still run on smaller GPUs: it is a mixture-of-experts model, so only about 3B of its 35.51B parameters are used for each generated token. Runtimes such as llama.cpp (--n-cpu-moe), ExLlamaV3 or Project Maya can keep most expert weights in system RAM and only the shared layers and hot experts in VRAM. Then you need VRAM + system RAM ≥ weights, and speed is limited mainly by RAM bandwidth.

FAQ

How much VRAM does Qwen3.6 35B-A3B need?

The weights are about 21.7 GB at Q4_K_M (35.51B params × 4.89 bits ÷ 8, before context/KV cache). Add memory for context (KV cache) and the runtime. The smallest-memory setup with a measured result here, fully in memory, is RTX 4090 (24 GB) at 160.6 tok/s.

What is the fastest measured hardware for Qwen3.6 35B-A3B?

MacBook Pro M5 Max 128 GB: Qwen3.6 35B-A3B at 271 tok/s (mlx-serve + MTP), Gemma 4 26B-A4B at 166: 271 tok/s generation (mlx-serve 26.10.2 + MTP, 4-bit MLX (mlx-serve build)), from https://baobabtech.ai/posts/local-moe-m5-max-qwen-gemma-speed.

Which GPUs have measured results for Qwen3.6 35B-A3B?

2x RTX 3090, DGX Spark (GB10), Intel Arc Pro B70, RTX 3090, RTX 4090, RTX 5090, RTX Spark N1X, Radeon AI Pro R9700, Radeon RX 7900 XTX, Strix Halo (Radeon 8060S).

Can it run on… 2x RTX 3090 · DGX Spark (GB10) · Intel Arc Pro B70 · RTX 3090 · RTX 4090 · RTX 5090 · RTX Spark N1X · Radeon AI Pro R9700 · Radeon RX 7900 XTX · Strix Halo (Radeon 8060S)