How much VRAM does MiniMax-M2 need?

Q4_K_M is exactly 138,342,383,904 bytes (128.84 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is MiniMax-M2's KV cache?

7.75 GiB at 32K context with an f16 cache, computed per layer from the model's architecture. Quantizing the cache to q8_0 roughly halves it.

Is MiniMax-M2 a mixture-of-experts model?

Yes — 256 experts with 8 routed per token. Every expert must be resident in memory, but only the routed ones are read per token, so its memory requirement and its speed behave very differently.

MiniMaxAI · text · mixture of experts

MiniMax-M2

MiniMaxAI/MiniMax-M2

MiniMax-M2 at Q4_K_M is exactly 138,342,383,904 bytes (128.84 GiB / 138.34 GB) — an effective 4.839 bits per weight, not the nominal 4. Its KV cache at 32K is 7.75 GiB.

From the file· summed from 4 file(s)From the file· KV per layer

Parameters

229B

total, not active

Architecture

minimax-m2

62 layers

Context

196,608

native (config.json)

License

mit

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ1_S	43.78 GiB	47,009,105,120	1.644	—	bartowski
IQ1_M	45.65 GiB	49,018,340,576	1.715	—	bartowski
IQ2_XXS2 shards	50.97 GiB	54,726,362,528	1.914	—	bartowski
UD-TQ1_0	52.50 GiB	56,369,923,264	1.972	—	unsloth
IQ2_XS2 shards	58.81 GiB	63,143,151,008	2.209	—	bartowski
IQ2_S2 shards	59.00 GiB	63,351,847,360	2.216	—	bartowski
UD-IQ1_S2 shards	59.58 GiB	63,976,293,760	2.238	—	unsloth
UD-IQ1_M2 shards	64.01 GiB	68,731,897,216	2.404	—	unsloth
IQ2_M2 shards	67.05 GiB	71,995,128,224	2.518	—	bartowski
UD-IQ2_XXS2 shards	68.92 GiB	74,000,401,792	2.588	—	unsloth
UD-IQ2_M2 shards	72.72 GiB	78,077,265,280	2.731	—	unsloth
Q2_K2 shards	74.34 GiB	79,821,736,384	2.792	—	bartowski
Q2_K_L3 shards	74.90 GiB	80,421,928,480	2.813	—	bartowski
Q2_K2 shards	77.58 GiB	83,302,681,984	2.914	—	unsloth
Q2_K_L2 shards	77.72 GiB	83,446,728,064	2.919	—	unsloth
IQ3_XXS3 shards	83.91 GiB	90,098,792,992	3.152	—	bartowski
UD-IQ3_XXS2 shards	87.18 GiB	93,608,314,240	3.274	—	unsloth
IQ3_XS3 shards	87.32 GiB	93,764,189,760	3.280	—	bartowski
Q3_K_S2 shards	91.91 GiB	98,690,597,248	3.452	—	unsloth
Q3_K_S3 shards	92.31 GiB	99,115,712,032	3.467	—	bartowski
IQ3_M3 shards	96.82 GiB	103,954,578,976	3.636	—	bartowski
Q3_K_M3 shards	96.82 GiB	103,955,365,408	3.636	—	bartowski
Q3_K_L3 shards	100.78 GiB	108,207,111,680	3.785	—	bartowski
Q3_K_M3 shards	101.77 GiB	109,272,482,272	3.822	—	unsloth
IQ4_XS3 shards	113.53 GiB	121,899,124,224	4.264	—	unsloth
IQ4_XS4 shards	113.78 GiB	122,166,375,040	4.273	—	bartowski
IQ4_NL3 shards	120.14 GiB	128,999,622,144	4.512	—	unsloth
IQ4_NL4 shards	120.37 GiB	129,243,415,200	4.521	—	bartowski
Q4_03 shards	120.61 GiB	129,503,725,056	4.530	—	unsloth
Q4_K_S3 shards	121.10 GiB	130,033,780,224	4.549	—	unsloth
Q4_04 shards	122.32 GiB	131,337,683,616	4.594	—	bartowski
Q4_K_S4 shards	124.56 GiB	133,750,456,992	4.679	—	bartowski
Q4_K_M4 shards	128.84 GiB	138,342,383,904	4.839	—	lmstudio-community
Q4_K_M3 shards	128.84 GiB	138,342,385,120	4.839	—	unsloth
Q4_K_M4 shards	129.07 GiB	138,586,178,208	4.848	—	bartowski
Q4_13 shards	133.39 GiB	143,226,840,576	5.010	—	unsloth
Q4_14 shards	133.47 GiB	143,312,167,584	5.013	—	bartowski
Q5_K_S4 shards	146.66 GiB	157,478,438,496	5.509	—	unsloth
Q5_K_S4 shards	146.73 GiB	157,551,575,712	5.511	—	bartowski
Q5_K_M4 shards	151.16 GiB	162,303,936,128	5.677	—	unsloth

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.97 GiB	0.97 GiB	—	62 / 0 / 0
8,192	1.94 GiB	1.94 GiB	—	62 / 0 / 0
16,384	3.88 GiB	3.88 GiB	—	62 / 0 / 0
32,768	7.75 GiB	7.75 GiB	—	62 / 0 / 0
65,536	15.50 GiB	15.50 GiB	—	62 / 0 / 0
131,072	31.00 GiB	31.00 GiB	—	62 / 0 / 0

Compare with

same modality, comparable size

vs Qwen3-Coder-30B-A3B-Instruct vs Qwen3.6-27B vs Qwen3.8-27B vs DeepSeek-V4-Flash vs Qwen3-30B-A3B-Thinking-2507 vs Laguna-XS-2.1 All alternatives →Will it fit on my card? →

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 119.81 GiB. The real file is 128.84 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json

Layers

Attention heads

KV heads

Head dim

128

Hidden size

3072

Vocab

200,064

Sliding window

none

SWA period

—

MLA

Experts

256

Experts per token

use_sliding_window

—

Questions people ask

How much VRAM does MiniMax-M2 need?: Q4_K_M is exactly 138,342,383,904 bytes (128.84 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is MiniMax-M2's KV cache?: 7.75 GiB at 32K context with an f16 cache, computed per layer. Quantizing the cache to q8_0 roughly halves it, which is often the difference between a context length fitting and not.
Is MiniMax-M2 a mixture-of-experts model?: Yes — 256 experts, 8 routed per token. Every expert must be resident, but only the routed ones are read per token, which is why its memory requirement and its speed behave very differently.
Which quantization of MiniMax-M2 should I use?: Q4_K_M is the usual default. Pick the largest quantization that fits your card at the context you actually need — the table above gives exact sizes for every one published.