How much VRAM does MiniCPM-V-4 need?

Q4_K_M is exactly 2,189,861,216 bytes (2.04 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is MiniCPM-V-4's KV cache?

1.00 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.

openbmb · vision language

MiniCPM-V-4

openbmb/MiniCPM-V-4

MiniCPM-V-4 at Q4_K_M is exactly 2,189,861,216 bytes (2.04 GiB / 2.19 GB) — an effective 4.316 bits per weight, not the nominal 4. Its KV cache at 32K is 1.00 GiB.

From the file· summed from 1 file(s)From the file· KV per layer

Parameters

4.1B

Architecture

llama

32 layers

Context

32,768

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
Q4_0	1.94 GiB	2,079,023,456	4.097	—	openbmb
Q4_K_S	1.95 GiB	2,092,458,336	4.123	—	openbmb
Q4_K_M	2.04 GiB	2,189,861,216	4.316	—	openbmb
Q4_1	2.14 GiB	2,292,626,016	4.518	—	openbmb
Q5_0	2.33 GiB	2,506,228,576	4.939	—	openbmb
Q5_K_S	2.33 GiB	2,506,228,576	4.939	—	openbmb
Q5_K_M	2.39 GiB	2,563,326,816	5.051	—	openbmb
Q5_1	2.53 GiB	2,719,831,136	5.360	—	openbmb
Q6_K	2.76 GiB	2,960,134,016	5.833	—	openbmb
Q8_0	3.57 GiB	3,833,381,696	7.554	—	openbmb

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.13 GiB	0.13 GiB	—	32 / 0 / 0
8,192	0.25 GiB	0.25 GiB	—	32 / 0 / 0
16,384	0.50 GiB	0.50 GiB	—	32 / 0 / 0
32,768	1.00 GiB	1.00 GiB	—	32 / 0 / 0
65,536	2.00 GiB	2.00 GiB	—	32 / 0 / 0
131,072	4.00 GiB	4.00 GiB	—	32 / 0 / 0

Compare with

same modality, comparable size

vs Qwen3.5-9B vs gemma-4-12B-it vs Qwen3.5-4B vs Qwen3.5-0.8B vs gemma-4-E2B-it vs gemma-4-E4B-it-qat-q4_0-unquantized 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 2.13 GiB. The real file is 2.04 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

2560

Vocab

73,448

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does MiniCPM-V-4 need?: Q4_K_M is exactly 2,189,861,216 bytes (2.04 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is MiniCPM-V-4's KV cache?: 1.00 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.
Which quantization of MiniCPM-V-4 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.