Watts

Name: OpenWeights Terminal measured pairs
Creator: Dvidia
License: https://owterminal.com/cite

Decode ÷ named TDP. No TDP, no row. Not a composite score.

#	Model · Build	Hardware	Decode	TDP	tok/100W	Hours / 1M
1	Qwen3.6-35B-A3B	M4 Pro	85.5	45W	190	3.25
2	Qwen3.5-4B	M4 Pro	82.8	45W	184	3.35
3	Qwen3.8-Flash-Next	M5 Max	126.5	70W	180.7	2.2
4	Qwen3.5-9B	M4 Pro	49.3	45W	109.6	5.63
5	Qwen3-8B	M4 Pro	48.3	45W	107.3	5.75
6	AAliceAI-Foundation-80B-A3B	M5 Max 128GB	65	70W	92.9	4.27
7	Qwen3.5 2B	RTX 5090 32GB	351	575W	61	0.79
8	Qwen3.6-35B-A3B	2× DGX Spark	94.4	240W	39.3	2.94
9	Qwen3.8-Flash-Next	DGX Spark	79.5	240W	33.1	3.49
10	MiMo-V2.6-Distill-Qwen-9B	RTX 5060 8GB	47	145W	32.4	5.91
11	OOrnith	GTX 1660 SUPER 6GB	40	125W	32	6.94
12	Gemma-4 26B-A4B-it	RTX 5090 32GB	173	575W	30.1	1.61
13	Empero Qwen3.8-35B-A3B	RTX 3060 12GB	50	170W	29.4	5.56
14	Qwen3.8-Flash-Next	2× DGX Spark	45	240W	18.8	6.17
15	Qwen3.8-Flash-Next	RTX 5090 32GB	80.9	575W	14.1	3.43
16	Qwen3.8-27B	RTX 5060 Ti 16GB	23	180W	12.8	12.08
17	Qwen3.8-27B	RTX 5060 Ti 16GB	19	180W	10.6	14.62
18	Qwen3.8-27B	RTX 4090 24GB	40.7	450W	9	6.83
19	NNex-N2.5-mini	RTX 5060 Ti 16GB	14	180W	7.8	19.84
20	Qwen3.8-27B TurboFCFusion	RTX 5060 Ti 16GB	10	180W	5.6	27.78