Qwen3.6-35B-A3B-FP8

Qwen · 36.0B (active NoneB) · MoE · ctx NoneK · apache-2.0 · HuggingFace

Qwen3.6-35B-A3B-FP8 (Qwen) is a 36.0B parameters MoE (36.0B active per token) open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 28.0 GB VRAM for Q4 (weights 19.8 GB + KV 7.2 GB + 1 GB overhead) and 80.2 GB for FP16. About 24 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 141.4 tok/s (estimated); MI300X 192GB ≈ 93.7 tok/s (estimated); H200 141GB SXM5 ≈ 84.8 tok/s (estimated); H100 80GB SXM5 ≈ 59.2 tok/s (estimated); Gaudi 3 128GB ≈ 57.9 tok/s (estimated). Use the table below for VRAM fit and tok/s per dollar (list MSRP). Estimates follow memory-bandwidth math; measured rows override when present. See methodology for details. Methodology.

92 hardware configs — sorted by tok/s. Default fit context 8K.

Hardware	VRAM	Quant	tok/s	Fits?	tok/s/$
B200 192GB SXM	192	Q4	141.4 est.	Yes	0.004
MI300X 192GB	192	Q4	93.7 est.	Yes	0.006
H200 141GB SXM5	141	Q4	84.8 est.	Yes	0.002
H100 80GB SXM5	80	Q4	59.2 est.	Yes	0.002
Gaudi 3 128GB	128	Q4	57.9 est.	Yes	0.005
H100 80GB SXM5	80	FP16	46.3 est.	No	0.002
B200 192GB SXM	192	FP16	38.9 est.	Yes	0.001
RTX 4090 24GB	24	Q4	37.3 est.	No	0.023
RTX 4090D 24GB (China)	24	Q4	37.3 est.	No	0.023
RTX 3090 Ti 24GB	24	Q4	37.3 est.	No	0.019
A100 80GB SXM4	80	Q4	36.0 est.	Yes	0.002
RX 7900 XTX 24GB	24	Q4	35.5 est.	No	0.036
RTX 3090 24GB	24	Q4	34.7 est.	No	0.023
RTX 5090 32GB	32	Q4	31.7 est.	Yes	0.016
A100 80GB SXM4	80	FP16	28.2 est.	No	0.002
MI300X 192GB	192	FP16	25.8 est.	Yes	0.002
H200 141GB SXM5	141	FP16	23.3 est.	Yes	0.001
RX 7900 XT 20GB	20	Q4	20.6 est.	No	0.023
RTX 6000 Ada 48GB	48	Q4	17.0 est.	Yes	0.003
V100 32GB SXM2	32	Q4	15.9 est.	Yes	0.002
Gaudi 3 128GB	128	FP16	15.9 est.	Yes	0.001
RTX 5080 16GB	16	Q4	15.8 est.	No	0.016
L40S 48GB	48	Q4	15.3 est.	Yes	0.002
V100 16GB PCIe	16	Q4	14.8 est.	No	0.002
RTX 5070 Ti 16GB	16	Q4	14.7 est.	No	0.02
Mac Studio M3 Ultra 96GB	96	Q4	14.5 est.	Yes	0.004
Mac Studio M2 Ultra 192GB	192	Q4	14.1 est.	Yes	0.002
RTX A6000 48GB	48	Q4	13.6 est.	Yes	0.003
RTX 4080 Super 16GB	16	Q4	12.1 est.	No	0.012
RTX 4070 Ti Super 16GB	16	Q4	11.0 est.	No	0.014
Arc Pro B70 32GB	32	Q4	10.7 est.	Yes	0.011
RX 9070 16GB	16	Q4	10.5 est.	No	0.019
RX 9070 XT 16GB	16	Q4	10.5 est.	No	0.018
RX 7800 XT 16GB	16	Q4	10.3 est.	No	0.021
MacBook Pro M4 Max 128GB	128	Q4	9.7 est.	Yes	0.002
RX 7900 GRE 16GB	16	Q4	9.5 est.	No	0.017
Arc A770 16GB	16	Q4	9.2 est.	No	0.026
DGX Spark (GH200 480GB)	480	Q4	8.8 est.	Yes	0.003
RTX 5060 Ti 16GB	16	Q4	7.4 est.	No	0.017
MacBook Pro M1 Max 32GB	32	Q4	7.1 est.	Yes	0.002
MacBook Pro M1 Max 64GB	64	Q4	7.1 est.	Yes	0.002
MacBook Pro M3 Max 128GB	128	Q4	7.1 est.	Yes	0.002
RTX 5070 12GB	12	Q4	6.2 est.	No	0.011
Titan V 12GB	12	Q4	6.0 est.	No	0.002
T4 16GB	16	Q4	5.3 est.	No	0.003
RTX 4070 Ti 12GB	12	Q4	5.1 est.	No	0.006
Titan Xp 12GB	12	Q4	5.1 est.	No	0.004
RTX 3080 10GB	10	Q4	4.9 est.	No	0.007
RTX 6000 Ada 48GB	48	FP16	4.8 est.	No	0.001
Mac mini M4 Pro 64GB	64	Q4	4.8 est.	Yes	0.002
RTX 4070 12GB	12	Q4	4.7 est.	No	0.008
RTX 4060 Ti 16GB	16	Q4	4.7 est.	No	0.009
RTX 4070 Super 12GB	12	Q4	4.7 est.	No	0.008
RX 7600 XT 16GB	16	Q4	4.7 est.	No	0.014
Titan X Pascal 12GB	12	Q4	4.4 est.	No	0.004
L40S 48GB	48	FP16	4.3 est.	No	0.001
Arc B580 12GB	12	Q4	4.2 est.	No	0.017
RTX 5090 32GB	32	FP16	4.0 est.	No	0.002
RX 7700 XT 12GB	12	Q4	4.0 est.	No	0.009
Mac Studio M3 Ultra 96GB	96	FP16	4.0 est.	Yes	0.001
Mac Studio M2 Ultra 192GB	192	FP16	3.9 est.	Yes	0.001
RTX A6000 48GB	48	FP16	3.8 est.	No	0.001
Ryzen AI Max+ 395 128GB	96	Q4	3.8 est.	Yes	0.002
Ryzen AI Max+ 395 64GB	48	Q4	3.8 est.	Yes	0.003
Ryzen AI Max 390 64GB	48	Q4	3.8 est.	Yes	0.003
MacBook Pro M1 Max 64GB	64	FP16	3.5 est.	No	0.001
RTX 3060 12GB	12	Q4	3.3 est.	No	0.01
MacBook Pro M4 Max 128GB	128	FP16	2.7 est.	Yes	0.001
DGX Spark (GH200 480GB)	480	FP16	2.4 est.	Yes	0.001
Arc B570 10GB	10	Q4	2.4 est.	No	0.011
Mac mini M4 Pro 64GB	64	FP16	2.4 est.	No	0.001
Mac mini M4 32GB	32	Q4	2.1 est.	Yes	0.002
V100 32GB SXM2	32	FP16	2.0 est.	No	0.0
MacBook Pro M3 Max 128GB	128	FP16	1.9 est.	Yes	0.0
RTX 5060 8GB	8	Q4	1.8 est.	No	0.006
RTX 5060 Ti 8GB	8	Q4	1.8 est.	No	0.005
RTX 3070 8GB	8	Q4	1.8 est.	No	0.004
RTX 4090 24GB	24	FP16	1.3 est.	No	0.001
RTX 4090D 24GB (China)	24	FP16	1.3 est.	No	0.001
RTX 3090 Ti 24GB	24	FP16	1.3 est.	No	0.001
Arc Pro B70 32GB	32	FP16	1.3 est.	No	0.001
RTX 3090 24GB	24	FP16	1.2 est.	No	0.001
RX 7900 XTX 24GB	24	FP16	1.2 est.	No	0.001
RTX 4060 Ti 8GB	8	Q4	1.2 est.	No	0.003
RX 7600 8GB	8	Q4	1.2 est.	No	0.004
RTX 4060 8GB	8	Q4	1.1 est.	No	0.004
Ryzen AI Max+ 395 64GB	48	FP16	1.1 est.	No	0.001
Ryzen AI Max 390 64GB	48	FP16	1.1 est.	No	0.001
Ryzen AI Max+ 395 128GB	96	FP16	1.0 est.	Yes	0.001
MacBook Pro M1 Max 32GB	32	FP16	0.9 est.	No	0.0
Tesla P4 8GB	8	Q4	0.8 est.	No	0.01
Mac mini M4 32GB	32	FP16	0.3 est.	No	0.0

Fit checks

• RTX 5090 32GB — fits · ≈31.7 tok/s Q4 est.
• RTX 3090 Ti 24GB — does not fit · ≈37.3 tok/s Q4 est.
• RTX 4090 24GB — does not fit · ≈37.3 tok/s Q4 est.
• RTX 5080 16GB — does not fit · ≈15.8 tok/s Q4 est.
• RTX 6000 Ada 48GB — fits · ≈17.0 tok/s Q4 est.
• RX 7900 XTX 24GB — does not fit · ≈35.5 tok/s Q4 est.
• RTX 3090 24GB — does not fit · ≈34.7 tok/s Q4 est.
• RTX 5070 Ti 16GB — does not fit · ≈14.7 tok/s Q4 est.
• L40S 48GB — fits · ≈15.3 tok/s Q4 est.
• Mac Studio M3 Ultra 96GB — fits · ≈14.5 tok/s Q4 est.
• Mac Studio M2 Ultra 192GB — fits · ≈14.1 tok/s Q4 est.
• RX 7900 XT 20GB — does not fit · ≈20.6 tok/s Q4 est.

12 popular retail GPUs; every row in the table above also links its Fits? verdict to the full fit check.

Same family: Qwen

Qwen3 Qwen3-8B · Qwen3-14B
Qwen2.5 Qwen2.5-7B-Instruct · Qwen2.5-32B-Instruct · Qwen2.5-14B-Instruct · Qwen2.5-72B-Instruct
Qwen3.5 Qwen3.5-9B · Qwen3.5-35B-A3B · Qwen3.5-2B · Qwen3.5-27B · Qwen3.5-122B-A10B · Qwen3.5-397B-A17B
Qwen3.6 Qwen3.6-35B-A3B · Qwen3.6-27B
Qwen3 2507 Qwen3-4B-Instruct-2507 · Qwen3-30B-A3B-Instruct-2507 · Qwen3-4B-Thinking-2507 · Qwen3-235B-A22B-Instruct-2507
Qwen2.5 Coder Qwen2.5-Coder-32B-Instruct
Qwen3 Coder Qwen3-Coder-Next · Qwen3-Coder-480B-A35B-Instruct
Qwen3 Next Qwen3-Next-80B-A3B-Instruct

All Qwen models →