Alibaba Qwen3-Coder

Qwen3-Coder-Next hardware requirements

Fits on one Instinct MI355X OAM at BF16. For 1,000 tokens/s at 32k context you need 4 x H100 SXM5 80GB at $7.32/hr or 4 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at $2.00/hr.

Parameters

79.7B

Active per token

3.0B

mixture of experts

KV cache per token

24 KB

16-bit cache

Native precision

BF16

Rent Model card

Top options for Qwen3-Coder-Next

The strongest part each vendor offers for Qwen3-Coder-Next at BF16 with 32,768 tokens per user, rentable parts first. Each card is one replica, the fewest accelerators that hold the model, and the line it draws on the graph below. Select a card to plot it.

Tokens per second as users grow

Each line is one replica, the fewest accelerators that hold the model, serving more users at once. It climbs while memory bandwidth is the limit, flattens once compute is, and stops where memory runs out at 32,768 tokens per user. Hover a line to bring it forward.

1 x GB300 NVL72 GPU 288GB 1 x Instinct MI355X OAM 2 x Ascend 950DT 2 x Gaudi 2 96GB 1 x TPU v7 192GB 8 x Inferentia2

Memory fills at: 1 x GB300 NVL72 GPU 288GB 133 users, 1 x Instinct MI355X OAM 133 users, 2 x Ascend 950DT 132 users, 2 x Gaudi 2 96GB 27 users, 1 x TPU v7 192GB 28 users, 8 x Inferentia2 90 users. Roofline estimates at BF16.

Bare minimum

Fewest GPUs

1 x Instinct MI355X OAM

BF16, 4,096 tokens, one stream

Tokens/s: 959
Per hour: $5.99
Per million tokens: $1.74
Memory used: 58%

Rent on Daytona

Cheapest per hour

2 x RTX PRO 6000 Blackwell Max-Q Workstation Edition

BF16, 4,096 tokens, one stream, live price

Tokens/s: 406
Per hour: $1.00
Per million tokens: $0.68
Memory used: 87%

Rent on RunPod

Cheapest per million tokens

4 x RTX PRO 6000 Blackwell Max-Q Workstation Edition

BF16, 32,768 tokens, 32 streams

Tokens/s: 1.1k
Per hour: $2.00
Per million tokens: $0.49
Memory used: 51%

Rent on RunPod

Configurations

Every datacenter GPU in the catalog, sized for these settings. Change precision, context and concurrency and the whole page follows. Each point is one replica: the smallest tensor-parallel group of that GPU that holds the model.

Precision Context Streams

Vendor

Priced only

Priced configurations only; 67 more fit without a live price and appear in the table. Marker size grows with the tensor-parallel width. Roofline estimates.

#									Rent
1	Instinct MI455X	1	45 %	4.0k tok/s	167 tok/s	—	—	—	Specs →
2	Rubin SXM	1	68 %	3.6k tok/s	158 tok/s	—	—	—	Specs →
3	TPU 8t announced	1	90 %	~1.3k tok/s	46.9 tok/s	—	—	—	Specs →
4	TPU 8i announced	1	68 %	~1.6k tok/s	61.8 tok/s	—	—	—	Specs →
5	Instinct MI355X OAM	1	68 %	1.5k tok/s	57.5 tok/s	$5.99	$1.11	1.4 kW	Rent →
6	GB300 NVL72 GPU 288GB	1	68 %	1.5k tok/s	57.5 tok/s	$12.42	$2.31	—	Rent →
7	GB200 NVL72 GPU 186GB	2	52 %	2.8k tok/s	109 tok/s	$21.00	$2.06	—	Rent →
8	Instinct MI350X OAM	1	68 %	1.5k tok/s	57.5 tok/s	$6.16	$1.16	1.0 kW	Rent →
9	TPU v7 192GB	2	51 %	2.6k tok/s	101 tok/s	—	—	—	Specs →
10	Instinct MI430X OAM announced	1	45 %	~3.1k tok/s	167 tok/s	—	—	—	Specs →
11	B300 SXM 262GB	1	74 %	1.5k tok/s	57.5 tok/s	$4.99	$0.95	1.4 kW	Rent →
12	B200 SXM 180GB	2	54 %	2.7k tok/s	105 tok/s	$8.18	$0.84	2.0 kW	Rent →
13	Ascend 970 announced	1	68 %	2.2k tok/s	104 tok/s	—	—	—	Specs →
14	B100 SXM 192GB	2	51 %	2.6k tok/s	109 tok/s	—	—	1.4 kW	Specs →
15	Jalapeño announced	1	90 %	~2.1k tok/s	111 tok/s	—	—	0.7 kW	Specs →
16	Instinct MI325X OAM	1	76 %	1.0k tok/s	43.1 tok/s	$3.80	$1.02	1.0 kW	Rent →
17	Instinct MI300X 192GB	2	51 %	1.8k tok/s	72.4 tok/s	$4.78	$0.74	1.5 kW	Rent →
18	BR100	4	77 %	1.2k tok/s	41.4 tok/s	—	—	2.2 kW	Specs →
19	Ascend 960 announced	1	68 %	1.3k tok/s	69.0 tok/s	—	—	—	Specs →
20	H200 SXM 141GB	2	69 %	1.6k tok/s	65.6 tok/s	$4.96	$0.89	1.4 kW	Rent →
21	H100 SXM5 80GB	4	61 %	2.2k tok/s	86.7 tok/s	$7.32	$0.91	2.8 kW	Rent →
22	TPU v6e 32GB	8	79 %	2.3k tok/s	80.1 tok/s	—	—	—	Specs →
23	Data Center GPU Max 1550 128GB	2	76 %	1.1k tok/s	44.8 tok/s	—	—	1.2 kW	Specs →
24	H200 NVL 141GB	2	69 %	~1.5k tok/s	65.6 tok/s	$7.23	$1.36	1.2 kW	Rent →
25	H100 NVL 94GB	4	52 %	~2.4k tok/s	101 tok/s	$12.76	$1.47	1.6 kW	Rent →
26	Ascend 910C	2	76 %	1.1k tok/s	43.7 tok/s	—	—	1.2 kW	Specs →
27	H100 PCIe 80GB	4	61 %	1.4k tok/s	51.8 tok/s	$10.00	$2.00	1.4 kW	Rent →
28	H800 PCIe 80GB	4	61 %	1.4k tok/s	51.8 tok/s	—	—	1.4 kW	Specs →
29	Ascend 950DT	2	68 %	1.1k tok/s	54.6 tok/s	—	—	—	Specs →
30	RTX PRO 6000 Blackwell Workstation Edition	4	51 %	1.2k tok/s	46.4 tok/s	$7.20	$1.70	2.4 kW	Rent →
31	Ascend 950PR	2	76 %	568 tok/s	21.9 tok/s	—	—	1.8 kW	Specs →
32	MTT S5000	4	61 %	~1.1k tok/s	41.4 tok/s	—	—	—	Specs →
33	RTX PRO 6000 Blackwell Server Edition	4	51 %	1.1k tok/s	41.3 tok/s	$2.76	$0.71	2.4 kW	Rent →
34	Gaudi 3 128GB	2	76 %	1.4k tok/s	50.5 tok/s	—	—	1.8 kW	Specs →
35	TPU v5p 95GB	4	52 %	1.6k tok/s	71.6 tok/s	—	—	1.8 kW	Specs →
36	RTX PRO 6000 Blackwell Max-Q Workstation Edition	4	51 %	1.1k tok/s	46.4 tok/s	$2.00	$0.49	1.2 kW	Rent →
37	Cloud AI 100 Ultra	2	76 %	~219 tok/s	7.5 tok/s	—	—	0.3 kW	Specs →
38	Gaudi 2 96GB	4	51 %	1.4k tok/s	63.4 tok/s	$4.40	$0.85	2.4 kW	Rent →
39	Atlas 350	2	88 %	494 tok/s	19.1 tok/s	—	—	1.2 kW	Specs →
40	Data Center GPU Max 1100 48GB	8	52 %	1.6k tok/s	60.1 tok/s	—	—	2.4 kW	Specs →
41	GeForce RTX 5090 32GB	8	79 %	1.7k tok/s	87.6 tok/s	$9.52	$1.53	4.6 kW	Rent →
42	Ascend 910B	4	77 %	1.0k tok/s	41.4 tok/s	—	—	1.6 kW	Specs →
43	MI250X 128GB	2	76 %	884 tok/s	44.8 tok/s	—	—	1.1 kW	Specs →
44	RTX 6000 Ada 48GB	8	52 %	~1.3k tok/s	46.9 tok/s	$7.76	$1.71	2.4 kW	Rent →
45	Instinct MI250 128GB	2	76 %	865 tok/s	44.8 tok/s	—	—	1.0 kW	Specs →
46	L40S 48GB	8	52 %	1.2k tok/s	42.2 tok/s	$6.08	$1.47	2.8 kW	Rent →
47	A800 PCIe 80GB	4	61 %	1.1k tok/s	50.1 tok/s	—	—	1.2 kW	Specs →
48	A800 SXM4	4	61 %	1.1k tok/s	52.8 tok/s	—	—	1.6 kW	Specs →
49	A100 PCIe 80GB	4	61 %	1.1k tok/s	50.1 tok/s	$5.40	$1.36	1.2 kW	Rent →
50	A100 SXM4 80GB	4	61 %	1.1k tok/s	52.8 tok/s	$5.60	$1.36	1.6 kW	Rent →
51	A100 PCIe 40GB	8	62 %	1.8k tok/s	76.0 tok/s	—	—	2.0 kW	Specs →
52	A100 SXM4 40GB	8	62 %	1.8k tok/s	76.0 tok/s	$12.49	$1.94	3.2 kW	Rent →
53	GeForce RTX 5090 D V2 24GB 2 nodes	16	53 %	~2.8k tok/s	116 tok/s	—	—	9.2 kW	Specs →
54	RTX 5880 Ada 48GB	8	52 %	~1.2k tok/s	46.9 tok/s	—	—	2.3 kW	Specs →
55	TPU v4 32GB	8	79 %	1.4k tok/s	58.7 tok/s	—	—	—	Specs →
56	RTX 5000 Ada 32GB	8	79 %	~777 tok/s	28.2 tok/s	$6.64	$2.37	2.0 kW	Rent →
57	RTX PRO 5000 Blackwell 48GB	8	52 %	~1.5k tok/s	65.7 tok/s	$7.68	$1.40	2.4 kW	Rent →
58	RTX PRO 5000 Blackwell 72GB	4	68 %	~808 tok/s	34.8 tok/s	—	—	1.2 kW	Specs →
59	RTX PRO 4500 Blackwell Server Edition	8	79 %	973 tok/s	39.1 tok/s	—	—	1.3 kW	Specs →
60	RTX PRO 4500 Blackwell 32GB	8	79 %	1.1k tok/s	43.8 tok/s	$5.76	$1.51	1.6 kW	Rent →
61	Radeon AI PRO R9700 32GB	8	79 %	806 tok/s	31.3 tok/s	—	—	2.4 kW	Specs →
62	Inferentia2	8	79 %	977 tok/s	40.1 tok/s	—	—	—	Specs →
63	Trainium	8	79 %	977 tok/s	40.1 tok/s	—	—	—	Specs →
64	Instinct MI100 32GB	8	79 %	978 tok/s	60.1 tok/s	—	—	2.4 kW	Specs →
65	Arc Pro B70 32GB	8	79 %	~767 tok/s	29.7 tok/s	—	—	1.8 kW	Specs →
66	L40 48GB	8	52 %	1.0k tok/s	42.2 tok/s	$6.56	$1.81	2.4 kW	Rent →
67	Instinct MI210 PCIe	4	77 %	819 tok/s	42.4 tok/s	—	—	1.2 kW	Specs →
68	GeForce RTX 4090 24GB 2 nodes	16	53 %	1.9k tok/s	87.0 tok/s	$14.24	$2.05	7.2 kW	Rent →
69	A30 24GB 2 nodes	16	53 %	1.8k tok/s	80.5 tok/s	—	—	2.6 kW	Specs →
70	RTX 4500 Ada 24GB 2 nodes	16	53 %	~996 tok/s	37.3 tok/s	—	—	3.4 kW	Specs →
71	RTX A6000 48GB	8	52 %	884 tok/s	37.5 tok/s	$4.00	$1.26	2.4 kW	Rent →
72	A40 48GB	8	52 %	814 tok/s	34.0 tok/s	$4.72	$1.61	2.4 kW	Rent →
73	H20 141GB HBM3e	2	69 %	591 tok/s	54.6 tok/s	—	—	0.8 kW	Specs →
74	H20 96GB	4	51 %	1.1k tok/s	104 tok/s	—	—	1.6 kW	Specs →
75	RTX PRO 4000 Blackwell 24GB 2 nodes	16	53 %	~1.4k tok/s	58.0 tok/s	$9.12	$1.82	2.3 kW	Rent →
76	BI-V100 Tiangai 100	8	79 %	1.2k tok/s	58.7 tok/s	—	—	2.0 kW	Specs →
77	Atlas 300I Duo 48GB	8	52 %	527 tok/s	19.9 tok/s	—	—	1.2 kW	Specs →
78	Atlas 300I Duo 96GB	4	51 %	279 tok/s	10.6 tok/s	—	—	0.6 kW	Specs →
79	RTX A5500 24GB 2 nodes	16	53 %	~1.5k tok/s	66.2 tok/s	—	—	3.7 kW	Specs →
80	Wormhole n300s 2 nodes	16	53 %	1.2k tok/s	49.7 tok/s	—	—	4.8 kW	Specs →
81	V100S PCIe 32GB	8	79 %	1.1k tok/s	55.4 tok/s	$7.04	$1.80	2.0 kW	Rent →
82	ATOM-Max	4	77 %	537 tok/s	26.5 tok/s	—	—	1.4 kW	Specs →
83	A10 24GB 2 nodes	16	53 %	1.2k tok/s	51.8 tok/s	$17.63	$3.98	2.4 kW	Rent →
84	Radeon PRO W7900 Dual Slot 48GB	8	52 %	896 tok/s	42.2 tok/s	—	—	2.4 kW	Specs →
85	Radeon PRO W7900 48GB	8	52 %	896 tok/s	42.2 tok/s	—	—	2.4 kW	Specs →
86	Radeon RX 7900 XTX 24GB 2 nodes	16	53 %	1.7k tok/s	82.8 tok/s	—	—	5.7 kW	Specs →
87	L4 24GB 2 nodes	16	53 %	702 tok/s	25.9 tok/s	$9.44	$3.74	1.2 kW	Rent →
88	L20 48GB	8	52 %	887 tok/s	42.2 tok/s	—	—	2.2 kW	Specs →
89	Radeon AI PRO R9600D 32GB	8	79 %	682 tok/s	31.3 tok/s	—	—	1.2 kW	Specs →
90	Arc Pro B65 32GB	8	79 %	~657 tok/s	29.7 tok/s	—	—	1.6 kW	Specs →
91	Arc Pro B60 24GB 2 nodes	16	53 %	~943 tok/s	39.3 tok/s	—	—	3.2 kW	Specs →
92	L2 24GB 2 nodes	16	53 %	676 tok/s	25.9 tok/s	—	—	—	Specs →
93	RTX PRO 4000 Blackwell SFF 24GB 2 nodes	16	53 %	~900 tok/s	37.3 tok/s	—	—	1.1 kW	Specs →
94	MLU370-X8	8	52 %	657 tok/s	30.0 tok/s	—	—	2.0 kW	Specs →
95	Radeon PRO W7800 48GB	8	52 %	798 tok/s	42.2 tok/s	—	—	2.1 kW	Specs →
96	Radeon PRO W7800 32GB	8	79 %	616 tok/s	28.2 tok/s	—	—	2.1 kW	Specs →
97	GeForce RTX 3090 24GB 2 nodes	16	53 %	1.3k tok/s	80.7 tok/s	$8.00	$1.68	5.6 kW	Rent →
98	A10G 24GB 2 nodes	16	53 %	1.0k tok/s	51.8 tok/s	$16.10	$4.38	4.8 kW	Rent →
99	Atlas 300V Pro	8	52 %	264 tok/s	10.0 tok/s	—	—	—	Specs →
100	Atlas 300I Pro 2 nodes	16	53 %	466 tok/s	17.7 tok/s	—	—	—	Specs →
101	Atlas 300V 2 nodes	16	53 %	436 tok/s	17.7 tok/s	—	—	1.2 kW	Specs →
102	GeForce RTX 3090 Ti 24GB 2 nodes	16	53 %	985 tok/s	87.0 tok/s	$7.36	$2.08	7.2 kW	Rent →
103	Radeon PRO W6800 32GB	8	79 %	393 tok/s	25.0 tok/s	—	—	2.0 kW	Specs →
104	RTX A5000 24GB 2 nodes	16	53 %	706 tok/s	66.2 tok/s	$4.32	$1.70	3.7 kW	Rent →
105	Radeon PRO V710 28GB	8	90 %	322 tok/s	21.9 tok/s	—	—	1.3 kW	Specs →

By vendor

At BF16 the NVIDIA pick is 4 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at about 1.1k tokens/s for $2.00/hr; the AMD pick is 2 x Instinct MI300X 192GB at about 1.8k tokens/s for $4.78/hr. Per rental dollar NVIDIA delivers 1.53x the tokens of AMD here.

NVIDIA 51 parts fit

Best value: 4 x RTX PRO 6000 Blackwell Max-Q Workstation Edition, 1.1k tok/s for $2.00/hr

Fewest GPUs: 1 x Rubin SXM, 3.6k tok/s

AMD 18 parts fit

Best value: 2 x Instinct MI300X 192GB, 1.8k tok/s for $4.78/hr

Fewest GPUs: 1 x Instinct MI455X, 4.0k tok/s

Huawei 10 parts fit

Fewest GPUs: 2 x Ascend 950DT, 1.1k tok/s

Intel 7 parts fit

Best value: 4 x Gaudi 2 96GB, 1.4k tok/s for $4.40/hr

Fewest GPUs: 2 x Gaudi 3 128GB, 1.4k tok/s

Google 4 parts fit

Fewest GPUs: 2 x TPU v7 192GB, 2.6k tok/s

AWS 2 parts fit

Fewest GPUs: 8 x Inferentia2, 977 tok/s

Other 7 parts fit

Fewest GPUs: 2 x Cloud AI 100 Ultra, 219 tok/s

Fleet what-if

Compare whole fleets at these settings: a thousand of one part against a hundred of another. Each fleet splits into replicas of its tensor-parallel width; GPUs left over sit idle.

A GPUs Part

B GPUs Part

Fleet letter	Fleet	Replicas	Idle	Tokens/s	Per hour	Power	Per M tokens
A	1,000 x H200 SXM 141GB	500 x TP2	0	776k tok/s	$2,480	700 kW	$0.89
B	100 x GB200 NVL72 GPU 186GB	50 x TP2	0	142k tok/s	$1,050	—	$2.06

Size Qwen3-Coder-Next for a tokens per second target

The calculator starts from your traffic instead of a fleet: set a target rate and read the replica count for every GPU.

Open in the LLM GPU Calculator

Frequently asked questions

How much GPU memory does Qwen3-Coder-Next need?

The weights take 159 GB at the native BF16 precision (159 GB at BF16, 79.7 GB at FP8, 44.8 GB at INT4). Each concurrent stream adds 24 KB of KV cache per token: 0.1 GB at 4,096 tokens and 0.8 GB at 32,768 tokens.

What is the bare minimum to run Qwen3-Coder-Next?

1 x Instinct MI355X OAM at BF16 with 4,096 tokens of context and one stream. The cheapest live rental for that footprint is 2 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at $1.00 per hour.

How many H100s do you need to run Qwen3-Coder-Next?

4 H100s at BF16 for 4,096 tokens of context and one stream. For 32,768 tokens and 32 concurrent streams the tensor-parallel group is 4 H100s, and 1,000 tokens per second takes 4 H100s across 1 replicas.

How much does Qwen3-Coder-Next cost per million tokens on an H100?

About $0.91 per million output tokens at BF16, 32,768 tokens of context and 32 streams, using the cheapest live on-demand price of $1.83 per GPU-hour. The cheapest part per token is 4 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at $0.49.

What precision does Qwen3-Coder-Next ship in?

The published checkpoint is BF16. FP8 and INT4 figures describe post-training quantisations that halve and quarter the footprint at a small accuracy cost.

All figures are roofline estimates from the model's config.json and each GPU's published memory, bandwidth and tensor-core peak, with fixed efficiency factors. Prices are the cheapest live on-demand listing per GPU when the site has one. They are a planning floor; a tuned serving stack can do better. Qwen3-Next hybrid: 12 of 48 layers keep a per-token cache and 36 linear-attention layers hold a fixed 75 MB of FP32 state per sequence. Active parameters per the Qwen model card.