MiniMax MiniMax M2

MiniMax M2.5 hardware requirements

Fits on one Instinct MI355X OAM at FP8. For 1,000 tokens/s at 32k context you need 8 x H100 SXM5 80GB at $14.64/hr or 16 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at $8.00/hr.

Parameters

229B

Active per token

10.0B

mixture of experts

KV cache per token

248 KB

16-bit cache

Native precision

FP8

Rent Model card

Top options for MiniMax M2.5

The strongest part each vendor offers for MiniMax M2.5 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 GB200 NVL72 GPU 186GB 1 x Instinct MI355X OAM 2 x Ascend 950DT 4 x Gaudi 2 96GB 2 x TPU v7 192GB 8 x Inferentia2

Memory fills at: 2 x GB200 NVL72 GPU 186GB 15 users, 1 x Instinct MI355X OAM 5 users, 2 x Ascend 950DT 5 users, 4 x Gaudi 2 96GB 16 users, 2 x TPU v7 192GB 16 users, 8 x Inferentia2 1 users. Roofline estimates at FP8.

Bare minimum

Fewest GPUs

1 x Instinct MI355X OAM

FP8, 4,096 tokens, one stream

Tokens/s: 537
Per hour: $5.99
Per million tokens: $3.10
Memory used: 83%

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: 431
Per hour: $2.00
Per million tokens: $1.29
Memory used: 63%

Rent on RunPod

Cheapest per million tokens

8 x RTX PRO 6000 Blackwell Max-Q Workstation Edition

FP8, 32,768 tokens, 32 streams

Tokens/s: 629
Per hour: $4.00
Per million tokens: $1.77
Memory used: 67%

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; 42 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	2	59 %	2.4k tok/s	80.3 tok/s	—	—	—	Specs →
2	Rubin SXM	2	89 %	2.3k tok/s	75.9 tok/s	—	—	—	Specs →
3	TPU 8t announced	4	60 %	~1.2k tok/s	42.7 tok/s	—	—	—	Specs →
4	TPU 8i announced	2	89 %	~786 tok/s	29.7 tok/s	—	—	—	Specs →
5	Instinct MI355X OAM	2	89 %	805 tok/s	27.6 tok/s	$11.98	$4.14	2.8 kW	Rent →
6	GB300 NVL72 GPU 288GB	2	89 %	804 tok/s	27.6 tok/s	$24.84	$8.58	—	Rent →
7	GB200 NVL72 GPU 186GB	4	69 %	1.5k tok/s	52.3 tok/s	$42.00	$7.66	—	Rent →
8	Instinct MI350X OAM	2	89 %	798 tok/s	27.6 tok/s	$12.32	$4.29	2.0 kW	Rent →
9	TPU v7 192GB	4	67 %	1.4k tok/s	48.2 tok/s	—	—	—	Specs →
10	Instinct MI430X OAM announced	2	59 %	~1.6k tok/s	80.3 tok/s	—	—	—	Specs →
11	B300 SXM 262GB	2	98 %	796 tok/s	27.6 tok/s	$9.98	$3.48	2.8 kW	Rent →
12	B200 SXM 180GB	4	72 %	1.5k tok/s	50.3 tok/s	$16.36	$3.12	4.0 kW	Rent →
13	Ascend 970 announced	2	89 %	1.3k tok/s	49.7 tok/s	—	—	—	Specs →
14	B100 SXM 192GB	4	67 %	1.5k tok/s	52.3 tok/s	—	—	2.8 kW	Specs →
15	Jalapeño announced	4	60 %	~2.5k tok/s	101 tok/s	—	—	2.8 kW	Specs →
16	Instinct MI325X OAM	4	50 %	1.1k tok/s	39.2 tok/s	$15.20	$3.84	4.0 kW	Rent →
17	Instinct MI300X 192GB	4	67 %	986 tok/s	34.6 tok/s	$9.56	$2.69	3.0 kW	Rent →
18	BR100 2 nodes	16	51 %	~1.0k tok/s	34.8 tok/s	—	—	8.8 kW	Specs →
19	Ascend 960 announced	2	89 %	819 tok/s	33.1 tok/s	—	—	—	Specs →
20	H200 SXM 141GB	4	92 %	874 tok/s	31.4 tok/s	$9.92	$3.15	2.8 kW	Rent →
21	H100 SXM5 80GB	8	81 %	1.2k tok/s	41.3 tok/s	$14.64	$3.39	5.6 kW	Rent →
22	Data Center GPU Max 1550 128GB	8	50 %	~1.0k tok/s	40.4 tok/s	—	—	4.8 kW	Specs →
23	H200 NVL 141GB	4	92 %	~854 tok/s	31.4 tok/s	$14.46	$4.70	2.4 kW	Rent →
24	H100 NVL 94GB	8	69 %	~1.3k tok/s	48.1 tok/s	$25.52	$5.26	3.2 kW	Rent →
25	Ascend 910C	8	50 %	~1.0k tok/s	39.5 tok/s	—	—	4.8 kW	Specs →
26	H100 PCIe 80GB	8	81 %	731 tok/s	24.7 tok/s	$20.00	$7.60	2.8 kW	Rent →
27	H800 PCIe 80GB	8	81 %	731 tok/s	24.7 tok/s	—	—	2.8 kW	Specs →
28	Ascend 950DT	4	90 %	677 tok/s	26.1 tok/s	—	—	—	Specs →
29	RTX PRO 6000 Blackwell Workstation Edition	8	67 %	638 tok/s	22.1 tok/s	$14.40	$6.27	4.8 kW	Rent →
30	Ascend 950PR	8	50 %	576 tok/s	19.7 tok/s	—	—	7.2 kW	Specs →
31	MTT S5000	8	81 %	~576 tok/s	19.7 tok/s	—	—	—	Specs →
32	RTX PRO 6000 Blackwell Server Edition	8	67 %	575 tok/s	19.7 tok/s	$5.52	$2.67	4.8 kW	Rent →
33	Gaudi 3 128GB	8	50 %	1.3k tok/s	45.7 tok/s	—	—	7.2 kW	Specs →
34	TPU v5p 95GB	8	68 %	799 tok/s	34.1 tok/s	—	—	3.6 kW	Specs →
35	RTX PRO 6000 Blackwell Max-Q Workstation Edition	8	67 %	629 tok/s	22.1 tok/s	$4.00	$1.77	2.4 kW	Rent →
36	Cloud AI 100 Ultra	8	50 %	~201 tok/s	6.8 tok/s	—	—	1.2 kW	Specs →
37	Gaudi 2 96GB	8	67 %	825 tok/s	30.2 tok/s	$8.80	$2.96	4.8 kW	Rent →
38	Atlas 350	8	58 %	500 tok/s	17.3 tok/s	—	—	4.8 kW	Specs →
39	Data Center GPU Max 1100 48GB 2 nodes	16	68 %	~727 tok/s	26.8 tok/s	—	—	4.8 kW	Specs →
40	Ascend 910B 2 nodes	16	51 %	~896 tok/s	34.8 tok/s	—	—	6.4 kW	Specs →
41	MI250X 128GB	8	50 %	~850 tok/s	40.4 tok/s	—	—	4.5 kW	Specs →
42	RTX 6000 Ada 48GB 2 nodes	16	68 %	~619 tok/s	20.9 tok/s	$15.52	$6.96	4.8 kW	Rent →
43	Instinct MI250 128GB	8	50 %	~834 tok/s	40.4 tok/s	—	—	4.0 kW	Specs →
44	L40S 48GB 2 nodes	16	68 %	562 tok/s	18.8 tok/s	$12.16	$6.01	5.6 kW	Rent →
45	A800 PCIe 80GB	8	81 %	~554 tok/s	23.9 tok/s	—	—	2.4 kW	Specs →
46	A800 SXM4	8	81 %	~576 tok/s	25.2 tok/s	—	—	3.2 kW	Specs →
47	A100 PCIe 80GB	8	81 %	~554 tok/s	23.9 tok/s	$10.80	$5.41	2.4 kW	Rent →
48	A100 SXM4 80GB	8	81 %	~576 tok/s	25.2 tok/s	$11.20	$5.40	3.2 kW	Rent →
49	A100 PCIe 40GB 2 nodes	16	82 %	~831 tok/s	33.9 tok/s	—	—	4.0 kW	Specs →
50	A100 SXM4 40GB 2 nodes	16	82 %	~831 tok/s	33.9 tok/s	$24.98	$8.35	6.4 kW	Rent →
51	RTX 5880 Ada 48GB 2 nodes	16	68 %	~605 tok/s	20.9 tok/s	—	—	4.6 kW	Specs →
52	RTX PRO 5000 Blackwell 48GB 2 nodes	16	68 %	~711 tok/s	29.3 tok/s	$15.36	$6.00	4.8 kW	Rent →
53	RTX PRO 5000 Blackwell 72GB	8	90 %	~403 tok/s	16.6 tok/s	—	—	2.4 kW	Specs →
54	L40 48GB 2 nodes	16	68 %	526 tok/s	18.8 tok/s	$13.12	$6.93	4.8 kW	Rent →
55	Instinct MI210 PCIe 2 nodes	16	51 %	~736 tok/s	35.7 tok/s	—	—	4.8 kW	Specs →
56	RTX A6000 48GB 2 nodes	16	68 %	~411 tok/s	16.7 tok/s	$8.00	$5.41	4.8 kW	Rent →
57	A40 48GB 2 nodes	16	68 %	~378 tok/s	15.2 tok/s	$9.44	$6.94	4.8 kW	Rent →
58	H20 141GB HBM3e	4	92 %	458 tok/s	26.1 tok/s	—	—	1.6 kW	Specs →
59	H20 96GB	8	67 %	866 tok/s	49.4 tok/s	—	—	3.2 kW	Specs →
60	Atlas 300I Duo 48GB 2 nodes	16	68 %	~241 tok/s	8.9 tok/s	—	—	2.4 kW	Specs →
61	Atlas 300I Duo 96GB	8	67 %	~137 tok/s	5.0 tok/s	—	—	1.2 kW	Specs →
62	ATOM-Max 2 nodes	16	51 %	~480 tok/s	22.3 tok/s	—	—	5.6 kW	Specs →
63	Radeon PRO W7900 Dual Slot 48GB 2 nodes	16	68 %	~422 tok/s	18.8 tok/s	—	—	4.7 kW	Specs →
64	Radeon PRO W7900 48GB 2 nodes	16	68 %	~422 tok/s	18.8 tok/s	—	—	4.7 kW	Specs →
65	L20 48GB 2 nodes	16	68 %	493 tok/s	18.8 tok/s	—	—	4.4 kW	Specs →
66	MLU370-X8 2 nodes	16	68 %	~308 tok/s	13.4 tok/s	—	—	4.0 kW	Specs →
67	Radeon PRO W7800 48GB 2 nodes	16	68 %	~380 tok/s	18.8 tok/s	—	—	4.2 kW	Specs →
68	Atlas 300V Pro 2 nodes	16	68 %	~121 tok/s	4.5 tok/s	—	—	—	Specs →

By vendor

At FP8 the NVIDIA pick is 8 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at about 629 tokens/s for $4.00/hr; the AMD pick is 4 x Instinct MI300X 192GB at about 986 tokens/s for $9.56/hr. Per rental dollar NVIDIA delivers 1.53x the tokens of AMD here.

NVIDIA 32 parts fit

Best value: 8 x RTX PRO 6000 Blackwell Max-Q Workstation Edition, 629 tok/s for $4.00/hr

Fewest GPUs: 2 x Rubin SXM, 2.3k tok/s

AMD 11 parts fit

Best value: 4 x Instinct MI300X 192GB, 986 tok/s for $9.56/hr

Fewest GPUs: 2 x Instinct MI455X, 2.4k tok/s

Huawei 8 parts fit

Fewest GPUs: 4 x Ascend 950DT, 677 tok/s

Intel 4 parts fit

Best value: 8 x Gaudi 2 96GB, 825 tok/s for $8.80/hr

Fewest GPUs: 8 x Gaudi 3 128GB, 1.3k tok/s

Google 2 parts fit

Fewest GPUs: 4 x TPU v7 192GB, 1.4k tok/s

Other 5 parts fit

Fewest GPUs: 8 x MTT S5000, 576 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	219k tok/s	$2,480	700 kW	$3.15
B	100 x GB200 NVL72 GPU 186GB	25 x TP4	0	38k tok/s	$1,050	—	$7.66

Size MiniMax M2.5 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 MiniMax M2.5 need?

The weights take 229 GB at the native FP8 precision (457 GB at BF16, 229 GB at FP8, 129 GB at INT4). Each concurrent stream adds 248 KB of KV cache per token: 1.0 GB at 4,096 tokens and 8.3 GB at 32,768 tokens.

What is the bare minimum to run MiniMax M2.5?

1 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 MiniMax M2.5?

4 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 MiniMax M2.5 cost per million tokens on an H100?

About $3.39 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 8 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at $1.77.

What precision does MiniMax M2.5 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. Same geometry as MiniMax M2, whose card states 10B active of 230B. Ships in FP8.