Xiaomi MiMo V2.5

MiMo-V2.5 hardware requirements

Needs at least 2 GPUs at FP8 (Instinct MI355X OAM). For 1,000 tokens/s at 32k context you need 8 x H100 SXM5 80GB at $14.64/hr or 8 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at $4.00/hr.

Parameters

311B

Active per token

15.0B

mixture of experts

KV cache per token

218 KB

16-bit cache

Native precision

FP8

Rent Model card

Top options for MiMo-V2.5

The strongest part each vendor offers for MiMo-V2.5 at FP8 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.

2 x GB300 NVL72 GPU 288GB 2 x Instinct MI355X OAM 4 x Ascend 950DT 4 x Gaudi 2 96GB 2 x TPU v7 192GB 16 x Inferentia2

Memory fills at: 2 x GB300 NVL72 GPU 288GB 312 users, 2 x Instinct MI355X OAM 312 users, 4 x Ascend 950DT 309 users, 4 x Gaudi 2 96GB 71 users, 2 x TPU v7 192GB 74 users, 16 x Inferentia2 213 users. Roofline estimates at FP8.

Bare minimum

Fewest GPUs

2 x Instinct MI355X OAM

FP8, 4,096 tokens, one stream

Tokens/s: 744
Per hour: $11.98
Per million tokens: $4.47
Memory used: 56%

Rent on Daytona

Cheapest per hour

4 x RTX PRO 6000 Blackwell Max-Q Workstation Edition

FP8, 4,096 tokens, one stream, live price

Tokens/s: 314
Per hour: $2.00
Per million tokens: $1.77
Memory used: 85%

Rent on RunPod

Cheapest per million tokens

4 x RTX PRO 6000 Blackwell Max-Q Workstation Edition

FP8, 32,768 tokens, 32 streams

Tokens/s: 512
Per hour: $2.00
Per million tokens: $1.08
Memory used: 92%

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	81 %	2.1k tok/s	77.6 tok/s	—	—	—	Specs →
2	Rubin SXM	2	61 %	3.7k tok/s	139 tok/s	—	—	—	Specs →
3	TPU 8t announced	2	81 %	~981 tok/s	41.3 tok/s	—	—	—	Specs →
4	TPU 8i announced	2	61 %	~1.1k tok/s	54.4 tok/s	—	—	—	Specs →
5	Instinct MI355X OAM	2	61 %	1.3k tok/s	50.6 tok/s	$11.98	$2.60	2.8 kW	Rent →
6	GB300 NVL72 GPU 288GB	2	61 %	1.3k tok/s	50.6 tok/s	$24.84	$5.40	—	Rent →
7	GB200 NVL72 GPU 186GB	2	94 %	1.3k tok/s	50.6 tok/s	$21.00	$4.57	—	Rent →
8	Instinct MI350X OAM	2	61 %	1.3k tok/s	50.6 tok/s	$12.32	$2.73	2.0 kW	Rent →
9	TPU v7 192GB	2	91 %	1.2k tok/s	46.7 tok/s	—	—	—	Specs →
10	Instinct MI430X OAM announced	1	81 %	~920 tok/s	77.6 tok/s	—	—	—	Specs →
11	B300 SXM 262GB	2	66 %	1.2k tok/s	50.6 tok/s	$9.98	$2.22	2.8 kW	Rent →
12	B200 SXM 180GB	2	97 %	1.2k tok/s	48.7 tok/s	$8.18	$1.88	2.0 kW	Rent →
13	Ascend 970 announced	2	61 %	1.8k tok/s	91.1 tok/s	—	—	—	Specs →
14	B100 SXM 192GB	2	91 %	1.2k tok/s	50.6 tok/s	—	—	1.4 kW	Specs →
15	Jalapeño announced	2	81 %	~1.8k tok/s	97.4 tok/s	—	—	1.4 kW	Specs →
16	Instinct MI325X OAM	2	68 %	877 tok/s	38.0 tok/s	$7.60	$2.41	2.0 kW	Rent →
17	Instinct MI300X 192GB	2	91 %	801 tok/s	33.5 tok/s	$4.78	$1.66	1.5 kW	Rent →
18	BR100	8	69 %	~919 tok/s	36.2 tok/s	—	—	4.4 kW	Specs →
19	Ascend 960 announced	2	61 %	1.1k tok/s	60.7 tok/s	—	—	—	Specs →
20	H200 SXM 141GB	4	62 %	1.3k tok/s	57.5 tok/s	$9.92	$2.11	2.8 kW	Rent →
21	H100 SXM5 80GB	8	55 %	1.9k tok/s	75.9 tok/s	$14.64	$2.15	5.6 kW	Rent →
22	TPU v6e 32GB 2 nodes	16	70 %	~1.6k tok/s	65.5 tok/s	—	—	—	Specs →
23	Data Center GPU Max 1550 128GB	4	68 %	~759 tok/s	39.3 tok/s	—	—	2.4 kW	Specs →
24	H200 NVL 141GB	4	62 %	~1.2k tok/s	57.5 tok/s	$14.46	$3.23	2.4 kW	Rent →
25	H100 NVL 94GB	4	94 %	~1.1k tok/s	46.8 tok/s	$12.76	$3.30	1.6 kW	Rent →
26	Ascend 910C	4	68 %	~735 tok/s	38.4 tok/s	—	—	2.4 kW	Specs →
27	H100 PCIe 80GB	8	55 %	1.2k tok/s	45.3 tok/s	$20.00	$4.68	2.8 kW	Rent →
28	H800 PCIe 80GB	8	55 %	1.2k tok/s	45.3 tok/s	—	—	2.8 kW	Specs →
29	Ascend 950DT	4	61 %	930 tok/s	48.0 tok/s	—	—	—	Specs →
30	RTX PRO 6000 Blackwell Workstation Edition	4	92 %	529 tok/s	21.5 tok/s	$7.20	$3.78	2.4 kW	Rent →
31	Ascend 950PR	4	68 %	484 tok/s	19.2 tok/s	—	—	3.6 kW	Specs →
32	MTT S5000	8	55 %	~914 tok/s	36.2 tok/s	—	—	—	Specs →
33	RTX PRO 6000 Blackwell Server Edition	4	92 %	483 tok/s	19.1 tok/s	$2.76	$1.59	2.4 kW	Rent →
34	Gaudi 3 128GB	4	68 %	1.0k tok/s	44.4 tok/s	—	—	3.6 kW	Specs →
35	TPU v5p 95GB	4	93 %	527 tok/s	33.1 tok/s	—	—	1.8 kW	Specs →
36	RTX PRO 6000 Blackwell Max-Q Workstation Edition	4	92 %	512 tok/s	21.5 tok/s	$2.00	$1.08	1.2 kW	Rent →
37	Cloud AI 100 Ultra	4	68 %	~174 tok/s	6.6 tok/s	—	—	0.6 kW	Specs →
38	Gaudi 2 96GB	4	92 %	637 tok/s	29.4 tok/s	$4.40	$1.92	2.4 kW	Rent →
39	Atlas 350	4	78 %	416 tok/s	16.8 tok/s	—	—	2.4 kW	Specs →
40	Data Center GPU Max 1100 48GB	8	93 %	~597 tok/s	27.8 tok/s	—	—	2.4 kW	Specs →
41	GeForce RTX 5090 32GB 2 nodes	16	70 %	1.7k tok/s	71.6 tok/s	$19.04	$3.14	9.2 kW	Rent →
42	Ascend 910B	8	69 %	~694 tok/s	36.2 tok/s	—	—	3.2 kW	Specs →
43	MI250X 128GB	4	68 %	~516 tok/s	39.3 tok/s	—	—	2.2 kW	Specs →
44	RTX 6000 Ada 48GB	8	93 %	~570 tok/s	21.7 tok/s	$7.76	$3.78	2.4 kW	Rent →
45	Instinct MI250 128GB	4	68 %	~499 tok/s	39.3 tok/s	—	—	2.0 kW	Specs →
46	L40S 48GB	8	93 %	523 tok/s	19.6 tok/s	$6.08	$3.23	2.8 kW	Rent →
47	A800 PCIe 80GB	8	55 %	~687 tok/s	43.8 tok/s	—	—	2.4 kW	Specs →
48	A800 SXM4	8	55 %	~705 tok/s	46.2 tok/s	—	—	3.2 kW	Specs →
49	A100 PCIe 80GB	8	55 %	~687 tok/s	43.8 tok/s	$10.80	$4.37	2.4 kW	Rent →
50	A100 SXM4 80GB	8	55 %	~705 tok/s	46.2 tok/s	$11.20	$4.42	3.2 kW	Rent →
51	A100 PCIe 40GB 2 nodes	16	56 %	~1.1k tok/s	62.1 tok/s	—	—	4.0 kW	Specs →
52	A100 SXM4 40GB 2 nodes	16	56 %	~1.1k tok/s	62.1 tok/s	$24.98	$6.41	6.4 kW	Rent →
53	GeForce RTX 5090 D V2 24GB 2 nodes	16	95 %	~977 tok/s	53.7 tok/s	—	—	9.2 kW	Specs →
54	RTX 5880 Ada 48GB	8	93 %	~539 tok/s	21.7 tok/s	—	—	2.3 kW	Specs →
55	TPU v4 32GB 2 nodes	16	70 %	~886 tok/s	48.0 tok/s	—	—	—	Specs →
56	RTX 5000 Ada 32GB 2 nodes	16	70 %	~622 tok/s	23.0 tok/s	$13.28	$5.94	4.0 kW	Rent →
57	RTX PRO 5000 Blackwell 48GB	8	93 %	~520 tok/s	30.4 tok/s	$7.68	$4.11	2.4 kW	Rent →
58	RTX PRO 5000 Blackwell 72GB	8	61 %	~520 tok/s	30.4 tok/s	—	—	2.4 kW	Specs →
59	RTX PRO 4500 Blackwell Server Edition 2 nodes	16	70 %	769 tok/s	32.0 tok/s	—	—	2.6 kW	Specs →
60	RTX PRO 4500 Blackwell 32GB 2 nodes	16	70 %	836 tok/s	35.8 tok/s	$11.52	$3.83	3.2 kW	Rent →
61	Radeon AI PRO R9700 32GB 2 nodes	16	70 %	639 tok/s	25.6 tok/s	—	—	4.8 kW	Specs →
62	Inferentia2 2 nodes	16	70 %	608 tok/s	32.8 tok/s	—	—	—	Specs →
63	Trainium 2 nodes	16	70 %	608 tok/s	32.8 tok/s	—	—	—	Specs →
64	Instinct MI100 32GB 2 nodes	16	70 %	~486 tok/s	49.1 tok/s	—	—	4.8 kW	Specs →
65	Arc Pro B70 32GB 2 nodes	16	70 %	~500 tok/s	24.3 tok/s	—	—	3.7 kW	Specs →
66	L40 48GB	8	93 %	447 tok/s	19.6 tok/s	$6.56	$4.08	2.4 kW	Rent →
67	Instinct MI210 PCIe	8	69 %	~471 tok/s	37.1 tok/s	—	—	2.4 kW	Specs →
68	GeForce RTX 4090 24GB 2 nodes	16	95 %	852 tok/s	40.3 tok/s	$14.24	$4.64	7.2 kW	Rent →
69	A30 24GB 2 nodes	16	95 %	~612 tok/s	37.3 tok/s	—	—	2.6 kW	Specs →
70	RTX 4500 Ada 24GB 2 nodes	16	95 %	~450 tok/s	17.3 tok/s	—	—	3.4 kW	Specs →
71	RTX A6000 48GB	8	93 %	~304 tok/s	17.4 tok/s	$4.00	$3.66	2.4 kW	Rent →
72	A40 48GB	8	93 %	~283 tok/s	15.8 tok/s	$4.72	$4.63	2.4 kW	Rent →
73	H20 141GB HBM3e	4	62 %	470 tok/s	48.0 tok/s	—	—	1.6 kW	Specs →
74	H20 96GB	4	92 %	470 tok/s	48.0 tok/s	—	—	1.6 kW	Specs →
75	RTX PRO 4000 Blackwell 24GB 2 nodes	16	95 %	~487 tok/s	26.9 tok/s	$9.12	$5.21	2.3 kW	Rent →
76	BI-V100 Tiangai 100 2 nodes	16	70 %	~647 tok/s	48.0 tok/s	—	—	4.0 kW	Specs →
77	Atlas 300I Duo 48GB	8	93 %	~199 tok/s	9.2 tok/s	—	—	1.2 kW	Specs →
78	Atlas 300I Duo 96GB	4	92 %	~105 tok/s	4.9 tok/s	—	—	0.6 kW	Specs →
79	RTX A5500 24GB 2 nodes	16	95 %	~505 tok/s	30.7 tok/s	—	—	3.7 kW	Specs →
80	Wormhole n300s 2 nodes	16	95 %	610 tok/s	23.0 tok/s	—	—	4.8 kW	Specs →
81	V100S PCIe 32GB 2 nodes	16	70 %	~589 tok/s	45.3 tok/s	$14.08	$6.65	4.0 kW	Rent →
82	ATOM-Max	8	69 %	~317 tok/s	23.2 tok/s	—	—	2.8 kW	Specs →
83	A10 24GB 2 nodes	16	95 %	~425 tok/s	24.0 tok/s	$17.63	$11.52	2.4 kW	Rent →
84	Radeon PRO W7900 Dual Slot 48GB	8	93 %	~287 tok/s	19.6 tok/s	—	—	2.4 kW	Specs →
85	Radeon PRO W7900 48GB	8	93 %	~287 tok/s	19.6 tok/s	—	—	2.4 kW	Specs →
86	Radeon RX 7900 XTX 24GB 2 nodes	16	95 %	~531 tok/s	38.4 tok/s	—	—	5.7 kW	Specs →
87	L4 24GB 2 nodes	16	95 %	318 tok/s	12.0 tok/s	$9.44	$8.25	1.2 kW	Rent →
88	L20 48GB	8	93 %	390 tok/s	19.6 tok/s	—	—	2.2 kW	Specs →
89	Radeon AI PRO R9600D 32GB 2 nodes	16	70 %	532 tok/s	25.6 tok/s	—	—	2.4 kW	Specs →
90	Arc Pro B65 32GB 2 nodes	16	70 %	~382 tok/s	24.3 tok/s	—	—	3.2 kW	Specs →
91	Arc Pro B60 24GB 2 nodes	16	95 %	~328 tok/s	18.2 tok/s	—	—	3.2 kW	Specs →
92	L2 24GB 2 nodes	16	95 %	~252 tok/s	12.0 tok/s	—	—	—	Specs →
93	RTX PRO 4000 Blackwell SFF 24GB 2 nodes	16	95 %	~315 tok/s	17.3 tok/s	—	—	1.1 kW	Specs →
94	MLU370-X8	8	93 %	~215 tok/s	13.9 tok/s	—	—	2.0 kW	Specs →
95	Radeon PRO W7800 48GB	8	93 %	~240 tok/s	19.6 tok/s	—	—	2.1 kW	Specs →
96	Radeon PRO W7800 32GB 2 nodes	16	70 %	~356 tok/s	23.0 tok/s	—	—	4.2 kW	Specs →
97	GeForce RTX 3090 24GB 2 nodes	16	95 %	~373 tok/s	37.4 tok/s	$8.00	$5.96	5.6 kW	Rent →
98	A10G 24GB 2 nodes	16	95 %	~315 tok/s	24.0 tok/s	$16.10	$14.21	4.8 kW	Rent →
99	Atlas 300V Pro	8	93 %	~99.5 tok/s	4.6 tok/s	—	—	—	Specs →
100	Atlas 300I Pro 2 nodes	16	95 %	~176 tok/s	8.2 tok/s	—	—	—	Specs →
101	Atlas 300V 2 nodes	16	95 %	~155 tok/s	8.2 tok/s	—	—	1.2 kW	Specs →
102	GeForce RTX 3090 Ti 24GB 2 nodes	16	95 %	~246 tok/s	40.3 tok/s	$7.36	$8.29	7.2 kW	Rent →
103	Radeon PRO W6800 32GB 2 nodes	16	70 %	~192 tok/s	20.5 tok/s	—	—	4.0 kW	Specs →
104	RTX A5000 24GB 2 nodes	16	95 %	~174 tok/s	30.7 tok/s	$4.32	$6.89	3.7 kW	Rent →
105	Radeon PRO V710 28GB 2 nodes	16	81 %	~154 tok/s	17.9 tok/s	—	—	2.5 kW	Specs →

By vendor

At FP8 the NVIDIA pick is 4 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at about 512 tokens/s for $2.00/hr; the AMD pick is 2 x Instinct MI300X 192GB at about 801 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, 512 tok/s for $2.00/hr

Fewest GPUs: 2 x Rubin SXM, 3.7k tok/s

AMD 18 parts fit

Best value: 2 x Instinct MI300X 192GB, 801 tok/s for $4.78/hr

Fewest GPUs: 1 x Instinct MI455X, 2.1k tok/s

Huawei 10 parts fit

Fewest GPUs: 4 x Ascend 950DT, 930 tok/s

Intel 7 parts fit

Best value: 4 x Gaudi 2 96GB, 637 tok/s for $4.40/hr

Fewest GPUs: 4 x Gaudi 3 128GB, 1.0k tok/s

Google 4 parts fit

Fewest GPUs: 2 x TPU v7 192GB, 1.2k tok/s

AWS 2 parts fit

Fewest GPUs: 16 x Inferentia2, 608 tok/s

Other 7 parts fit

Fewest GPUs: 4 x Cloud AI 100 Ultra, 174 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	250 x TP4	0	327k tok/s	$2,480	700 kW	$2.11
B	100 x GB200 NVL72 GPU 186GB	50 x TP2	0	64k tok/s	$1,050	—	$4.57

Size MiMo-V2.5 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 MiMo-V2.5 need?

The weights take 311 GB at the native FP8 precision (622 GB at BF16, 311 GB at FP8, 175 GB at INT4). Each concurrent stream adds 218 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 MiMo-V2.5?

2 x Instinct MI355X OAM at FP8 with 4,096 tokens of context and one stream. The cheapest live rental for that footprint is 4 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at $2.00 per hour.

How many H100s do you need to run MiMo-V2.5?

8 H100s at FP8 for 4,096 tokens of context and one stream. For 32,768 tokens and 32 concurrent streams the tensor-parallel group is 8 H100s, and 1,000 tokens per second takes 8 H100s across 1 replicas.

How much does MiMo-V2.5 cost per million tokens on an H100?

About $2.15 per million output tokens at FP8, 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 $1.08.

What precision does MiMo-V2.5 ship in?

The published checkpoint is FP8. FP8 halves the BF16 footprint with the accuracy the lab validated; INT4 is a further community quantisation.

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. 39 of 48 layers attend over a 128-token sliding window, so the cache grows only in the 9 full layers. Keys are 192 wide and values 128, counted exactly. Ships in FP8. Active parameters per the Xiaomi model card.